2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/aio.h>
41 #include <linux/bitops.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
53 struct ext4_inode_info
*ei
)
55 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
60 csum_lo
= le16_to_cpu(raw
->i_checksum_lo
);
61 raw
->i_checksum_lo
= 0;
62 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
63 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
64 csum_hi
= le16_to_cpu(raw
->i_checksum_hi
);
65 raw
->i_checksum_hi
= 0;
68 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
,
69 EXT4_INODE_SIZE(inode
->i_sb
));
71 raw
->i_checksum_lo
= cpu_to_le16(csum_lo
);
72 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
73 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
74 raw
->i_checksum_hi
= cpu_to_le16(csum_hi
);
79 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
80 struct ext4_inode_info
*ei
)
82 __u32 provided
, calculated
;
84 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
85 cpu_to_le32(EXT4_OS_LINUX
) ||
86 !ext4_has_metadata_csum(inode
->i_sb
))
89 provided
= le16_to_cpu(raw
->i_checksum_lo
);
90 calculated
= ext4_inode_csum(inode
, raw
, ei
);
91 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
92 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
93 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
97 return provided
== calculated
;
100 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
101 struct ext4_inode_info
*ei
)
105 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
106 cpu_to_le32(EXT4_OS_LINUX
) ||
107 !ext4_has_metadata_csum(inode
->i_sb
))
110 csum
= ext4_inode_csum(inode
, raw
, ei
);
111 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
112 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
113 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
114 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
120 trace_ext4_begin_ordered_truncate(inode
, new_size
);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode
)->jinode
)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
130 EXT4_I(inode
)->jinode
,
134 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
135 unsigned int length
);
136 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
137 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
138 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
142 * Test whether an inode is a fast symlink.
144 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
146 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
147 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
149 if (ext4_has_inline_data(inode
))
152 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
156 * Restart the transaction associated with *handle. This does a commit,
157 * so before we call here everything must be consistently dirtied against
160 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
166 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
167 * moment, get_block can be called only for blocks inside i_size since
168 * page cache has been already dropped and writes are blocked by
169 * i_mutex. So we can safely drop the i_data_sem here.
171 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
172 jbd_debug(2, "restarting handle %p\n", handle
);
173 up_write(&EXT4_I(inode
)->i_data_sem
);
174 ret
= ext4_journal_restart(handle
, nblocks
);
175 down_write(&EXT4_I(inode
)->i_data_sem
);
176 ext4_discard_preallocations(inode
);
182 * Called at the last iput() if i_nlink is zero.
184 void ext4_evict_inode(struct inode
*inode
)
189 trace_ext4_evict_inode(inode
);
191 if (inode
->i_nlink
) {
193 * When journalling data dirty buffers are tracked only in the
194 * journal. So although mm thinks everything is clean and
195 * ready for reaping the inode might still have some pages to
196 * write in the running transaction or waiting to be
197 * checkpointed. Thus calling jbd2_journal_invalidatepage()
198 * (via truncate_inode_pages()) to discard these buffers can
199 * cause data loss. Also even if we did not discard these
200 * buffers, we would have no way to find them after the inode
201 * is reaped and thus user could see stale data if he tries to
202 * read them before the transaction is checkpointed. So be
203 * careful and force everything to disk here... We use
204 * ei->i_datasync_tid to store the newest transaction
205 * containing inode's data.
207 * Note that directories do not have this problem because they
208 * don't use page cache.
210 if (ext4_should_journal_data(inode
) &&
211 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
212 inode
->i_ino
!= EXT4_JOURNAL_INO
) {
213 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
214 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
216 jbd2_complete_transaction(journal
, commit_tid
);
217 filemap_write_and_wait(&inode
->i_data
);
219 truncate_inode_pages_final(&inode
->i_data
);
221 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
225 if (is_bad_inode(inode
))
227 dquot_initialize(inode
);
229 if (ext4_should_order_data(inode
))
230 ext4_begin_ordered_truncate(inode
, 0);
231 truncate_inode_pages_final(&inode
->i_data
);
233 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
236 * Protect us against freezing - iput() caller didn't have to have any
237 * protection against it
239 sb_start_intwrite(inode
->i_sb
);
240 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
241 ext4_blocks_for_truncate(inode
)+3);
242 if (IS_ERR(handle
)) {
243 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
249 ext4_orphan_del(NULL
, inode
);
250 sb_end_intwrite(inode
->i_sb
);
255 ext4_handle_sync(handle
);
257 err
= ext4_mark_inode_dirty(handle
, inode
);
259 ext4_warning(inode
->i_sb
,
260 "couldn't mark inode dirty (err %d)", err
);
264 ext4_truncate(inode
);
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
272 if (!ext4_handle_has_enough_credits(handle
, 3)) {
273 err
= ext4_journal_extend(handle
, 3);
275 err
= ext4_journal_restart(handle
, 3);
277 ext4_warning(inode
->i_sb
,
278 "couldn't extend journal (err %d)", err
);
280 ext4_journal_stop(handle
);
281 ext4_orphan_del(NULL
, inode
);
282 sb_end_intwrite(inode
->i_sb
);
288 * Kill off the orphan record which ext4_truncate created.
289 * AKPM: I think this can be inside the above `if'.
290 * Note that ext4_orphan_del() has to be able to cope with the
291 * deletion of a non-existent orphan - this is because we don't
292 * know if ext4_truncate() actually created an orphan record.
293 * (Well, we could do this if we need to, but heck - it works)
295 ext4_orphan_del(handle
, inode
);
296 EXT4_I(inode
)->i_dtime
= get_seconds();
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
305 if (ext4_mark_inode_dirty(handle
, inode
))
306 /* If that failed, just do the required in-core inode clear. */
307 ext4_clear_inode(inode
);
309 ext4_free_inode(handle
, inode
);
310 ext4_journal_stop(handle
);
311 sb_end_intwrite(inode
->i_sb
);
314 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
318 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
320 return &EXT4_I(inode
)->i_reserved_quota
;
325 * Called with i_data_sem down, which is important since we can call
326 * ext4_discard_preallocations() from here.
328 void ext4_da_update_reserve_space(struct inode
*inode
,
329 int used
, int quota_claim
)
331 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
332 struct ext4_inode_info
*ei
= EXT4_I(inode
);
334 spin_lock(&ei
->i_block_reservation_lock
);
335 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
336 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
337 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
338 "with only %d reserved data blocks",
339 __func__
, inode
->i_ino
, used
,
340 ei
->i_reserved_data_blocks
);
342 used
= ei
->i_reserved_data_blocks
;
345 /* Update per-inode reservations */
346 ei
->i_reserved_data_blocks
-= used
;
347 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
349 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
351 /* Update quota subsystem for data blocks */
353 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
356 * We did fallocate with an offset that is already delayed
357 * allocated. So on delayed allocated writeback we should
358 * not re-claim the quota for fallocated blocks.
360 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
364 * If we have done all the pending block allocations and if
365 * there aren't any writers on the inode, we can discard the
366 * inode's preallocations.
368 if ((ei
->i_reserved_data_blocks
== 0) &&
369 (atomic_read(&inode
->i_writecount
) == 0))
370 ext4_discard_preallocations(inode
);
373 static int __check_block_validity(struct inode
*inode
, const char *func
,
375 struct ext4_map_blocks
*map
)
377 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
379 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
380 "lblock %lu mapped to illegal pblock "
381 "(length %d)", (unsigned long) map
->m_lblk
,
388 #define check_block_validity(inode, map) \
389 __check_block_validity((inode), __func__, __LINE__, (map))
391 #ifdef ES_AGGRESSIVE_TEST
392 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
394 struct ext4_map_blocks
*es_map
,
395 struct ext4_map_blocks
*map
,
402 * There is a race window that the result is not the same.
403 * e.g. xfstests #223 when dioread_nolock enables. The reason
404 * is that we lookup a block mapping in extent status tree with
405 * out taking i_data_sem. So at the time the unwritten extent
406 * could be converted.
408 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
409 down_read(&EXT4_I(inode
)->i_data_sem
);
410 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
411 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
412 EXT4_GET_BLOCKS_KEEP_SIZE
);
414 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
415 EXT4_GET_BLOCKS_KEEP_SIZE
);
417 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
418 up_read((&EXT4_I(inode
)->i_data_sem
));
421 * We don't check m_len because extent will be collpased in status
422 * tree. So the m_len might not equal.
424 if (es_map
->m_lblk
!= map
->m_lblk
||
425 es_map
->m_flags
!= map
->m_flags
||
426 es_map
->m_pblk
!= map
->m_pblk
) {
427 printk("ES cache assertion failed for inode: %lu "
428 "es_cached ex [%d/%d/%llu/%x] != "
429 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
430 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
431 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
432 map
->m_len
, map
->m_pblk
, map
->m_flags
,
436 #endif /* ES_AGGRESSIVE_TEST */
439 * The ext4_map_blocks() function tries to look up the requested blocks,
440 * and returns if the blocks are already mapped.
442 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
443 * and store the allocated blocks in the result buffer head and mark it
446 * If file type is extents based, it will call ext4_ext_map_blocks(),
447 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
450 * On success, it returns the number of blocks being mapped or allocated.
451 * if create==0 and the blocks are pre-allocated and unwritten block,
452 * the result buffer head is unmapped. If the create ==1, it will make sure
453 * the buffer head is mapped.
455 * It returns 0 if plain look up failed (blocks have not been allocated), in
456 * that case, buffer head is unmapped
458 * It returns the error in case of allocation failure.
460 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
461 struct ext4_map_blocks
*map
, int flags
)
463 struct extent_status es
;
466 #ifdef ES_AGGRESSIVE_TEST
467 struct ext4_map_blocks orig_map
;
469 memcpy(&orig_map
, map
, sizeof(*map
));
473 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
474 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
475 (unsigned long) map
->m_lblk
);
478 * ext4_map_blocks returns an int, and m_len is an unsigned int
480 if (unlikely(map
->m_len
> INT_MAX
))
481 map
->m_len
= INT_MAX
;
483 /* We can handle the block number less than EXT_MAX_BLOCKS */
484 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
487 /* Lookup extent status tree firstly */
488 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
489 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
490 map
->m_pblk
= ext4_es_pblock(&es
) +
491 map
->m_lblk
- es
.es_lblk
;
492 map
->m_flags
|= ext4_es_is_written(&es
) ?
493 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
494 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
495 if (retval
> map
->m_len
)
498 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
503 #ifdef ES_AGGRESSIVE_TEST
504 ext4_map_blocks_es_recheck(handle
, inode
, map
,
511 * Try to see if we can get the block without requesting a new
514 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
515 down_read(&EXT4_I(inode
)->i_data_sem
);
516 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
517 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
518 EXT4_GET_BLOCKS_KEEP_SIZE
);
520 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
521 EXT4_GET_BLOCKS_KEEP_SIZE
);
526 if (unlikely(retval
!= map
->m_len
)) {
527 ext4_warning(inode
->i_sb
,
528 "ES len assertion failed for inode "
529 "%lu: retval %d != map->m_len %d",
530 inode
->i_ino
, retval
, map
->m_len
);
534 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
535 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
536 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
537 ext4_find_delalloc_range(inode
, map
->m_lblk
,
538 map
->m_lblk
+ map
->m_len
- 1))
539 status
|= EXTENT_STATUS_DELAYED
;
540 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
541 map
->m_len
, map
->m_pblk
, status
);
545 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
546 up_read((&EXT4_I(inode
)->i_data_sem
));
549 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
550 ret
= check_block_validity(inode
, map
);
555 /* If it is only a block(s) look up */
556 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
560 * Returns if the blocks have already allocated
562 * Note that if blocks have been preallocated
563 * ext4_ext_get_block() returns the create = 0
564 * with buffer head unmapped.
566 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
568 * If we need to convert extent to unwritten
569 * we continue and do the actual work in
570 * ext4_ext_map_blocks()
572 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
576 * Here we clear m_flags because after allocating an new extent,
577 * it will be set again.
579 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
582 * New blocks allocate and/or writing to unwritten extent
583 * will possibly result in updating i_data, so we take
584 * the write lock of i_data_sem, and call get_block()
585 * with create == 1 flag.
587 down_write(&EXT4_I(inode
)->i_data_sem
);
590 * We need to check for EXT4 here because migrate
591 * could have changed the inode type in between
593 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
594 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
596 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
598 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
600 * We allocated new blocks which will result in
601 * i_data's format changing. Force the migrate
602 * to fail by clearing migrate flags
604 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
608 * Update reserved blocks/metadata blocks after successful
609 * block allocation which had been deferred till now. We don't
610 * support fallocate for non extent files. So we can update
611 * reserve space here.
614 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
615 ext4_da_update_reserve_space(inode
, retval
, 1);
621 if (unlikely(retval
!= map
->m_len
)) {
622 ext4_warning(inode
->i_sb
,
623 "ES len assertion failed for inode "
624 "%lu: retval %d != map->m_len %d",
625 inode
->i_ino
, retval
, map
->m_len
);
630 * If the extent has been zeroed out, we don't need to update
631 * extent status tree.
633 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
634 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
635 if (ext4_es_is_written(&es
))
638 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
639 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
640 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
641 ext4_find_delalloc_range(inode
, map
->m_lblk
,
642 map
->m_lblk
+ map
->m_len
- 1))
643 status
|= EXTENT_STATUS_DELAYED
;
644 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
645 map
->m_pblk
, status
);
651 up_write((&EXT4_I(inode
)->i_data_sem
));
652 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
653 ret
= check_block_validity(inode
, map
);
660 /* Maximum number of blocks we map for direct IO at once. */
661 #define DIO_MAX_BLOCKS 4096
663 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
664 struct buffer_head
*bh
, int flags
)
666 handle_t
*handle
= ext4_journal_current_handle();
667 struct ext4_map_blocks map
;
668 int ret
= 0, started
= 0;
671 if (ext4_has_inline_data(inode
))
675 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
677 if (flags
&& !(flags
& EXT4_GET_BLOCKS_NO_LOCK
) && !handle
) {
678 /* Direct IO write... */
679 if (map
.m_len
> DIO_MAX_BLOCKS
)
680 map
.m_len
= DIO_MAX_BLOCKS
;
681 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
682 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
684 if (IS_ERR(handle
)) {
685 ret
= PTR_ERR(handle
);
691 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
693 ext4_io_end_t
*io_end
= ext4_inode_aio(inode
);
695 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
696 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
697 if (io_end
&& io_end
->flag
& EXT4_IO_END_UNWRITTEN
)
698 set_buffer_defer_completion(bh
);
699 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
703 ext4_journal_stop(handle
);
707 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
708 struct buffer_head
*bh
, int create
)
710 return _ext4_get_block(inode
, iblock
, bh
,
711 create
? EXT4_GET_BLOCKS_CREATE
: 0);
715 * `handle' can be NULL if create is zero
717 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
718 ext4_lblk_t block
, int create
)
720 struct ext4_map_blocks map
;
721 struct buffer_head
*bh
;
724 J_ASSERT(handle
!= NULL
|| create
== 0);
728 err
= ext4_map_blocks(handle
, inode
, &map
,
729 create
? EXT4_GET_BLOCKS_CREATE
: 0);
732 return create
? ERR_PTR(-ENOSPC
) : NULL
;
736 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
738 return ERR_PTR(-ENOMEM
);
739 if (map
.m_flags
& EXT4_MAP_NEW
) {
740 J_ASSERT(create
!= 0);
741 J_ASSERT(handle
!= NULL
);
744 * Now that we do not always journal data, we should
745 * keep in mind whether this should always journal the
746 * new buffer as metadata. For now, regular file
747 * writes use ext4_get_block instead, so it's not a
751 BUFFER_TRACE(bh
, "call get_create_access");
752 err
= ext4_journal_get_create_access(handle
, bh
);
757 if (!buffer_uptodate(bh
)) {
758 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
759 set_buffer_uptodate(bh
);
762 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
763 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
767 BUFFER_TRACE(bh
, "not a new buffer");
774 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
775 ext4_lblk_t block
, int create
)
777 struct buffer_head
*bh
;
779 bh
= ext4_getblk(handle
, inode
, block
, create
);
782 if (!bh
|| buffer_uptodate(bh
))
784 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
786 if (buffer_uptodate(bh
))
789 return ERR_PTR(-EIO
);
792 int ext4_walk_page_buffers(handle_t
*handle
,
793 struct buffer_head
*head
,
797 int (*fn
)(handle_t
*handle
,
798 struct buffer_head
*bh
))
800 struct buffer_head
*bh
;
801 unsigned block_start
, block_end
;
802 unsigned blocksize
= head
->b_size
;
804 struct buffer_head
*next
;
806 for (bh
= head
, block_start
= 0;
807 ret
== 0 && (bh
!= head
|| !block_start
);
808 block_start
= block_end
, bh
= next
) {
809 next
= bh
->b_this_page
;
810 block_end
= block_start
+ blocksize
;
811 if (block_end
<= from
|| block_start
>= to
) {
812 if (partial
&& !buffer_uptodate(bh
))
816 err
= (*fn
)(handle
, bh
);
824 * To preserve ordering, it is essential that the hole instantiation and
825 * the data write be encapsulated in a single transaction. We cannot
826 * close off a transaction and start a new one between the ext4_get_block()
827 * and the commit_write(). So doing the jbd2_journal_start at the start of
828 * prepare_write() is the right place.
830 * Also, this function can nest inside ext4_writepage(). In that case, we
831 * *know* that ext4_writepage() has generated enough buffer credits to do the
832 * whole page. So we won't block on the journal in that case, which is good,
833 * because the caller may be PF_MEMALLOC.
835 * By accident, ext4 can be reentered when a transaction is open via
836 * quota file writes. If we were to commit the transaction while thus
837 * reentered, there can be a deadlock - we would be holding a quota
838 * lock, and the commit would never complete if another thread had a
839 * transaction open and was blocking on the quota lock - a ranking
842 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
843 * will _not_ run commit under these circumstances because handle->h_ref
844 * is elevated. We'll still have enough credits for the tiny quotafile
847 int do_journal_get_write_access(handle_t
*handle
,
848 struct buffer_head
*bh
)
850 int dirty
= buffer_dirty(bh
);
853 if (!buffer_mapped(bh
) || buffer_freed(bh
))
856 * __block_write_begin() could have dirtied some buffers. Clean
857 * the dirty bit as jbd2_journal_get_write_access() could complain
858 * otherwise about fs integrity issues. Setting of the dirty bit
859 * by __block_write_begin() isn't a real problem here as we clear
860 * the bit before releasing a page lock and thus writeback cannot
861 * ever write the buffer.
864 clear_buffer_dirty(bh
);
865 BUFFER_TRACE(bh
, "get write access");
866 ret
= ext4_journal_get_write_access(handle
, bh
);
868 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
872 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
873 struct buffer_head
*bh_result
, int create
);
874 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
875 loff_t pos
, unsigned len
, unsigned flags
,
876 struct page
**pagep
, void **fsdata
)
878 struct inode
*inode
= mapping
->host
;
879 int ret
, needed_blocks
;
886 trace_ext4_write_begin(inode
, pos
, len
, flags
);
888 * Reserve one block more for addition to orphan list in case
889 * we allocate blocks but write fails for some reason
891 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
892 index
= pos
>> PAGE_CACHE_SHIFT
;
893 from
= pos
& (PAGE_CACHE_SIZE
- 1);
896 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
897 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
906 * grab_cache_page_write_begin() can take a long time if the
907 * system is thrashing due to memory pressure, or if the page
908 * is being written back. So grab it first before we start
909 * the transaction handle. This also allows us to allocate
910 * the page (if needed) without using GFP_NOFS.
913 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
919 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
920 if (IS_ERR(handle
)) {
921 page_cache_release(page
);
922 return PTR_ERR(handle
);
926 if (page
->mapping
!= mapping
) {
927 /* The page got truncated from under us */
929 page_cache_release(page
);
930 ext4_journal_stop(handle
);
933 /* In case writeback began while the page was unlocked */
934 wait_for_stable_page(page
);
936 if (ext4_should_dioread_nolock(inode
))
937 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
939 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
941 if (!ret
&& ext4_should_journal_data(inode
)) {
942 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
944 do_journal_get_write_access
);
950 * __block_write_begin may have instantiated a few blocks
951 * outside i_size. Trim these off again. Don't need
952 * i_size_read because we hold i_mutex.
954 * Add inode to orphan list in case we crash before
957 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
958 ext4_orphan_add(handle
, inode
);
960 ext4_journal_stop(handle
);
961 if (pos
+ len
> inode
->i_size
) {
962 ext4_truncate_failed_write(inode
);
964 * If truncate failed early the inode might
965 * still be on the orphan list; we need to
966 * make sure the inode is removed from the
967 * orphan list in that case.
970 ext4_orphan_del(NULL
, inode
);
973 if (ret
== -ENOSPC
&&
974 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
976 page_cache_release(page
);
983 /* For write_end() in data=journal mode */
984 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
987 if (!buffer_mapped(bh
) || buffer_freed(bh
))
989 set_buffer_uptodate(bh
);
990 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
991 clear_buffer_meta(bh
);
992 clear_buffer_prio(bh
);
997 * We need to pick up the new inode size which generic_commit_write gave us
998 * `file' can be NULL - eg, when called from page_symlink().
1000 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1001 * buffers are managed internally.
1003 static int ext4_write_end(struct file
*file
,
1004 struct address_space
*mapping
,
1005 loff_t pos
, unsigned len
, unsigned copied
,
1006 struct page
*page
, void *fsdata
)
1008 handle_t
*handle
= ext4_journal_current_handle();
1009 struct inode
*inode
= mapping
->host
;
1011 int i_size_changed
= 0;
1013 trace_ext4_write_end(inode
, pos
, len
, copied
);
1014 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
)) {
1015 ret
= ext4_jbd2_file_inode(handle
, inode
);
1018 page_cache_release(page
);
1023 if (ext4_has_inline_data(inode
)) {
1024 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1030 copied
= block_write_end(file
, mapping
, pos
,
1031 len
, copied
, page
, fsdata
);
1033 * it's important to update i_size while still holding page lock:
1034 * page writeout could otherwise come in and zero beyond i_size.
1036 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1038 page_cache_release(page
);
1041 * Don't mark the inode dirty under page lock. First, it unnecessarily
1042 * makes the holding time of page lock longer. Second, it forces lock
1043 * ordering of page lock and transaction start for journaling
1047 ext4_mark_inode_dirty(handle
, inode
);
1049 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1050 /* if we have allocated more blocks and copied
1051 * less. We will have blocks allocated outside
1052 * inode->i_size. So truncate them
1054 ext4_orphan_add(handle
, inode
);
1056 ret2
= ext4_journal_stop(handle
);
1060 if (pos
+ len
> inode
->i_size
) {
1061 ext4_truncate_failed_write(inode
);
1063 * If truncate failed early the inode might still be
1064 * on the orphan list; we need to make sure the inode
1065 * is removed from the orphan list in that case.
1068 ext4_orphan_del(NULL
, inode
);
1071 return ret
? ret
: copied
;
1074 static int ext4_journalled_write_end(struct file
*file
,
1075 struct address_space
*mapping
,
1076 loff_t pos
, unsigned len
, unsigned copied
,
1077 struct page
*page
, void *fsdata
)
1079 handle_t
*handle
= ext4_journal_current_handle();
1080 struct inode
*inode
= mapping
->host
;
1084 int size_changed
= 0;
1086 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1087 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1090 BUG_ON(!ext4_handle_valid(handle
));
1092 if (ext4_has_inline_data(inode
))
1093 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1097 if (!PageUptodate(page
))
1099 page_zero_new_buffers(page
, from
+copied
, to
);
1102 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1103 to
, &partial
, write_end_fn
);
1105 SetPageUptodate(page
);
1107 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1108 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1109 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1111 page_cache_release(page
);
1114 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1119 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1120 /* if we have allocated more blocks and copied
1121 * less. We will have blocks allocated outside
1122 * inode->i_size. So truncate them
1124 ext4_orphan_add(handle
, inode
);
1126 ret2
= ext4_journal_stop(handle
);
1129 if (pos
+ len
> inode
->i_size
) {
1130 ext4_truncate_failed_write(inode
);
1132 * If truncate failed early the inode might still be
1133 * on the orphan list; we need to make sure the inode
1134 * is removed from the orphan list in that case.
1137 ext4_orphan_del(NULL
, inode
);
1140 return ret
? ret
: copied
;
1144 * Reserve a single cluster located at lblock
1146 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1148 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1149 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1150 unsigned int md_needed
;
1154 * We will charge metadata quota at writeout time; this saves
1155 * us from metadata over-estimation, though we may go over by
1156 * a small amount in the end. Here we just reserve for data.
1158 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1163 * recalculate the amount of metadata blocks to reserve
1164 * in order to allocate nrblocks
1165 * worse case is one extent per block
1167 spin_lock(&ei
->i_block_reservation_lock
);
1169 * ext4_calc_metadata_amount() has side effects, which we have
1170 * to be prepared undo if we fail to claim space.
1173 trace_ext4_da_reserve_space(inode
, 0);
1175 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1176 spin_unlock(&ei
->i_block_reservation_lock
);
1177 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1180 ei
->i_reserved_data_blocks
++;
1181 spin_unlock(&ei
->i_block_reservation_lock
);
1183 return 0; /* success */
1186 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1188 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1189 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1192 return; /* Nothing to release, exit */
1194 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1196 trace_ext4_da_release_space(inode
, to_free
);
1197 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1199 * if there aren't enough reserved blocks, then the
1200 * counter is messed up somewhere. Since this
1201 * function is called from invalidate page, it's
1202 * harmless to return without any action.
1204 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1205 "ino %lu, to_free %d with only %d reserved "
1206 "data blocks", inode
->i_ino
, to_free
,
1207 ei
->i_reserved_data_blocks
);
1209 to_free
= ei
->i_reserved_data_blocks
;
1211 ei
->i_reserved_data_blocks
-= to_free
;
1213 /* update fs dirty data blocks counter */
1214 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1216 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1218 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1221 static void ext4_da_page_release_reservation(struct page
*page
,
1222 unsigned int offset
,
1223 unsigned int length
)
1226 struct buffer_head
*head
, *bh
;
1227 unsigned int curr_off
= 0;
1228 struct inode
*inode
= page
->mapping
->host
;
1229 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1230 unsigned int stop
= offset
+ length
;
1234 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
1236 head
= page_buffers(page
);
1239 unsigned int next_off
= curr_off
+ bh
->b_size
;
1241 if (next_off
> stop
)
1244 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1246 clear_buffer_delay(bh
);
1248 curr_off
= next_off
;
1249 } while ((bh
= bh
->b_this_page
) != head
);
1252 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1253 ext4_es_remove_extent(inode
, lblk
, to_release
);
1256 /* If we have released all the blocks belonging to a cluster, then we
1257 * need to release the reserved space for that cluster. */
1258 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1259 while (num_clusters
> 0) {
1260 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1261 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1262 if (sbi
->s_cluster_ratio
== 1 ||
1263 !ext4_find_delalloc_cluster(inode
, lblk
))
1264 ext4_da_release_space(inode
, 1);
1271 * Delayed allocation stuff
1274 struct mpage_da_data
{
1275 struct inode
*inode
;
1276 struct writeback_control
*wbc
;
1278 pgoff_t first_page
; /* The first page to write */
1279 pgoff_t next_page
; /* Current page to examine */
1280 pgoff_t last_page
; /* Last page to examine */
1282 * Extent to map - this can be after first_page because that can be
1283 * fully mapped. We somewhat abuse m_flags to store whether the extent
1284 * is delalloc or unwritten.
1286 struct ext4_map_blocks map
;
1287 struct ext4_io_submit io_submit
; /* IO submission data */
1290 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1295 struct pagevec pvec
;
1296 struct inode
*inode
= mpd
->inode
;
1297 struct address_space
*mapping
= inode
->i_mapping
;
1299 /* This is necessary when next_page == 0. */
1300 if (mpd
->first_page
>= mpd
->next_page
)
1303 index
= mpd
->first_page
;
1304 end
= mpd
->next_page
- 1;
1306 ext4_lblk_t start
, last
;
1307 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1308 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1309 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1312 pagevec_init(&pvec
, 0);
1313 while (index
<= end
) {
1314 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1317 for (i
= 0; i
< nr_pages
; i
++) {
1318 struct page
*page
= pvec
.pages
[i
];
1319 if (page
->index
> end
)
1321 BUG_ON(!PageLocked(page
));
1322 BUG_ON(PageWriteback(page
));
1324 block_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
1325 ClearPageUptodate(page
);
1329 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1330 pagevec_release(&pvec
);
1334 static void ext4_print_free_blocks(struct inode
*inode
)
1336 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1337 struct super_block
*sb
= inode
->i_sb
;
1338 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1340 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1341 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1342 ext4_count_free_clusters(sb
)));
1343 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1344 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1345 (long long) EXT4_C2B(EXT4_SB(sb
),
1346 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1347 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1348 (long long) EXT4_C2B(EXT4_SB(sb
),
1349 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1350 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1351 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1352 ei
->i_reserved_data_blocks
);
1356 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1358 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1362 * This function is grabs code from the very beginning of
1363 * ext4_map_blocks, but assumes that the caller is from delayed write
1364 * time. This function looks up the requested blocks and sets the
1365 * buffer delay bit under the protection of i_data_sem.
1367 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1368 struct ext4_map_blocks
*map
,
1369 struct buffer_head
*bh
)
1371 struct extent_status es
;
1373 sector_t invalid_block
= ~((sector_t
) 0xffff);
1374 #ifdef ES_AGGRESSIVE_TEST
1375 struct ext4_map_blocks orig_map
;
1377 memcpy(&orig_map
, map
, sizeof(*map
));
1380 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1384 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1385 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1386 (unsigned long) map
->m_lblk
);
1388 /* Lookup extent status tree firstly */
1389 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1390 if (ext4_es_is_hole(&es
)) {
1392 down_read(&EXT4_I(inode
)->i_data_sem
);
1397 * Delayed extent could be allocated by fallocate.
1398 * So we need to check it.
1400 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1401 map_bh(bh
, inode
->i_sb
, invalid_block
);
1403 set_buffer_delay(bh
);
1407 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1408 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1409 if (retval
> map
->m_len
)
1410 retval
= map
->m_len
;
1411 map
->m_len
= retval
;
1412 if (ext4_es_is_written(&es
))
1413 map
->m_flags
|= EXT4_MAP_MAPPED
;
1414 else if (ext4_es_is_unwritten(&es
))
1415 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1419 #ifdef ES_AGGRESSIVE_TEST
1420 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1426 * Try to see if we can get the block without requesting a new
1427 * file system block.
1429 down_read(&EXT4_I(inode
)->i_data_sem
);
1430 if (ext4_has_inline_data(inode
))
1432 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1433 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1435 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1441 * XXX: __block_prepare_write() unmaps passed block,
1445 * If the block was allocated from previously allocated cluster,
1446 * then we don't need to reserve it again. However we still need
1447 * to reserve metadata for every block we're going to write.
1449 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
<= 1 ||
1450 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1451 ret
= ext4_da_reserve_space(inode
, iblock
);
1453 /* not enough space to reserve */
1459 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1460 ~0, EXTENT_STATUS_DELAYED
);
1466 map_bh(bh
, inode
->i_sb
, invalid_block
);
1468 set_buffer_delay(bh
);
1469 } else if (retval
> 0) {
1471 unsigned int status
;
1473 if (unlikely(retval
!= map
->m_len
)) {
1474 ext4_warning(inode
->i_sb
,
1475 "ES len assertion failed for inode "
1476 "%lu: retval %d != map->m_len %d",
1477 inode
->i_ino
, retval
, map
->m_len
);
1481 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1482 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1483 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1484 map
->m_pblk
, status
);
1490 up_read((&EXT4_I(inode
)->i_data_sem
));
1496 * This is a special get_block_t callback which is used by
1497 * ext4_da_write_begin(). It will either return mapped block or
1498 * reserve space for a single block.
1500 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1501 * We also have b_blocknr = -1 and b_bdev initialized properly
1503 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1504 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1505 * initialized properly.
1507 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1508 struct buffer_head
*bh
, int create
)
1510 struct ext4_map_blocks map
;
1513 BUG_ON(create
== 0);
1514 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1516 map
.m_lblk
= iblock
;
1520 * first, we need to know whether the block is allocated already
1521 * preallocated blocks are unmapped but should treated
1522 * the same as allocated blocks.
1524 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1528 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1529 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1531 if (buffer_unwritten(bh
)) {
1532 /* A delayed write to unwritten bh should be marked
1533 * new and mapped. Mapped ensures that we don't do
1534 * get_block multiple times when we write to the same
1535 * offset and new ensures that we do proper zero out
1536 * for partial write.
1539 set_buffer_mapped(bh
);
1544 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1550 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1556 static int __ext4_journalled_writepage(struct page
*page
,
1559 struct address_space
*mapping
= page
->mapping
;
1560 struct inode
*inode
= mapping
->host
;
1561 struct buffer_head
*page_bufs
= NULL
;
1562 handle_t
*handle
= NULL
;
1563 int ret
= 0, err
= 0;
1564 int inline_data
= ext4_has_inline_data(inode
);
1565 struct buffer_head
*inode_bh
= NULL
;
1567 ClearPageChecked(page
);
1570 BUG_ON(page
->index
!= 0);
1571 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1572 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1573 if (inode_bh
== NULL
)
1576 page_bufs
= page_buffers(page
);
1581 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1584 /* As soon as we unlock the page, it can go away, but we have
1585 * references to buffers so we are safe */
1588 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1589 ext4_writepage_trans_blocks(inode
));
1590 if (IS_ERR(handle
)) {
1591 ret
= PTR_ERR(handle
);
1595 BUG_ON(!ext4_handle_valid(handle
));
1598 BUFFER_TRACE(inode_bh
, "get write access");
1599 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1601 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1604 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1605 do_journal_get_write_access
);
1607 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1612 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1613 err
= ext4_journal_stop(handle
);
1617 if (!ext4_has_inline_data(inode
))
1618 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1620 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1627 * Note that we don't need to start a transaction unless we're journaling data
1628 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1629 * need to file the inode to the transaction's list in ordered mode because if
1630 * we are writing back data added by write(), the inode is already there and if
1631 * we are writing back data modified via mmap(), no one guarantees in which
1632 * transaction the data will hit the disk. In case we are journaling data, we
1633 * cannot start transaction directly because transaction start ranks above page
1634 * lock so we have to do some magic.
1636 * This function can get called via...
1637 * - ext4_writepages after taking page lock (have journal handle)
1638 * - journal_submit_inode_data_buffers (no journal handle)
1639 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1640 * - grab_page_cache when doing write_begin (have journal handle)
1642 * We don't do any block allocation in this function. If we have page with
1643 * multiple blocks we need to write those buffer_heads that are mapped. This
1644 * is important for mmaped based write. So if we do with blocksize 1K
1645 * truncate(f, 1024);
1646 * a = mmap(f, 0, 4096);
1648 * truncate(f, 4096);
1649 * we have in the page first buffer_head mapped via page_mkwrite call back
1650 * but other buffer_heads would be unmapped but dirty (dirty done via the
1651 * do_wp_page). So writepage should write the first block. If we modify
1652 * the mmap area beyond 1024 we will again get a page_fault and the
1653 * page_mkwrite callback will do the block allocation and mark the
1654 * buffer_heads mapped.
1656 * We redirty the page if we have any buffer_heads that is either delay or
1657 * unwritten in the page.
1659 * We can get recursively called as show below.
1661 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1664 * But since we don't do any block allocation we should not deadlock.
1665 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1667 static int ext4_writepage(struct page
*page
,
1668 struct writeback_control
*wbc
)
1673 struct buffer_head
*page_bufs
= NULL
;
1674 struct inode
*inode
= page
->mapping
->host
;
1675 struct ext4_io_submit io_submit
;
1676 bool keep_towrite
= false;
1678 trace_ext4_writepage(page
);
1679 size
= i_size_read(inode
);
1680 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1681 len
= size
& ~PAGE_CACHE_MASK
;
1683 len
= PAGE_CACHE_SIZE
;
1685 page_bufs
= page_buffers(page
);
1687 * We cannot do block allocation or other extent handling in this
1688 * function. If there are buffers needing that, we have to redirty
1689 * the page. But we may reach here when we do a journal commit via
1690 * journal_submit_inode_data_buffers() and in that case we must write
1691 * allocated buffers to achieve data=ordered mode guarantees.
1693 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
1694 ext4_bh_delay_or_unwritten
)) {
1695 redirty_page_for_writepage(wbc
, page
);
1696 if (current
->flags
& PF_MEMALLOC
) {
1698 * For memory cleaning there's no point in writing only
1699 * some buffers. So just bail out. Warn if we came here
1700 * from direct reclaim.
1702 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
1707 keep_towrite
= true;
1710 if (PageChecked(page
) && ext4_should_journal_data(inode
))
1712 * It's mmapped pagecache. Add buffers and journal it. There
1713 * doesn't seem much point in redirtying the page here.
1715 return __ext4_journalled_writepage(page
, len
);
1717 ext4_io_submit_init(&io_submit
, wbc
);
1718 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
1719 if (!io_submit
.io_end
) {
1720 redirty_page_for_writepage(wbc
, page
);
1724 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
1725 ext4_io_submit(&io_submit
);
1726 /* Drop io_end reference we got from init */
1727 ext4_put_io_end_defer(io_submit
.io_end
);
1731 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
1734 loff_t size
= i_size_read(mpd
->inode
);
1737 BUG_ON(page
->index
!= mpd
->first_page
);
1738 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1739 len
= size
& ~PAGE_CACHE_MASK
;
1741 len
= PAGE_CACHE_SIZE
;
1742 clear_page_dirty_for_io(page
);
1743 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
1745 mpd
->wbc
->nr_to_write
--;
1751 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1754 * mballoc gives us at most this number of blocks...
1755 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1756 * The rest of mballoc seems to handle chunks up to full group size.
1758 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1761 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1763 * @mpd - extent of blocks
1764 * @lblk - logical number of the block in the file
1765 * @bh - buffer head we want to add to the extent
1767 * The function is used to collect contig. blocks in the same state. If the
1768 * buffer doesn't require mapping for writeback and we haven't started the
1769 * extent of buffers to map yet, the function returns 'true' immediately - the
1770 * caller can write the buffer right away. Otherwise the function returns true
1771 * if the block has been added to the extent, false if the block couldn't be
1774 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
1775 struct buffer_head
*bh
)
1777 struct ext4_map_blocks
*map
= &mpd
->map
;
1779 /* Buffer that doesn't need mapping for writeback? */
1780 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
1781 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
1782 /* So far no extent to map => we write the buffer right away */
1783 if (map
->m_len
== 0)
1788 /* First block in the extent? */
1789 if (map
->m_len
== 0) {
1792 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
1796 /* Don't go larger than mballoc is willing to allocate */
1797 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
1800 /* Can we merge the block to our big extent? */
1801 if (lblk
== map
->m_lblk
+ map
->m_len
&&
1802 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
1810 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1812 * @mpd - extent of blocks for mapping
1813 * @head - the first buffer in the page
1814 * @bh - buffer we should start processing from
1815 * @lblk - logical number of the block in the file corresponding to @bh
1817 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1818 * the page for IO if all buffers in this page were mapped and there's no
1819 * accumulated extent of buffers to map or add buffers in the page to the
1820 * extent of buffers to map. The function returns 1 if the caller can continue
1821 * by processing the next page, 0 if it should stop adding buffers to the
1822 * extent to map because we cannot extend it anymore. It can also return value
1823 * < 0 in case of error during IO submission.
1825 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
1826 struct buffer_head
*head
,
1827 struct buffer_head
*bh
,
1830 struct inode
*inode
= mpd
->inode
;
1832 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
1833 >> inode
->i_blkbits
;
1836 BUG_ON(buffer_locked(bh
));
1838 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
1839 /* Found extent to map? */
1842 /* Everything mapped so far and we hit EOF */
1845 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
1846 /* So far everything mapped? Submit the page for IO. */
1847 if (mpd
->map
.m_len
== 0) {
1848 err
= mpage_submit_page(mpd
, head
->b_page
);
1852 return lblk
< blocks
;
1856 * mpage_map_buffers - update buffers corresponding to changed extent and
1857 * submit fully mapped pages for IO
1859 * @mpd - description of extent to map, on return next extent to map
1861 * Scan buffers corresponding to changed extent (we expect corresponding pages
1862 * to be already locked) and update buffer state according to new extent state.
1863 * We map delalloc buffers to their physical location, clear unwritten bits,
1864 * and mark buffers as uninit when we perform writes to unwritten extents
1865 * and do extent conversion after IO is finished. If the last page is not fully
1866 * mapped, we update @map to the next extent in the last page that needs
1867 * mapping. Otherwise we submit the page for IO.
1869 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
1871 struct pagevec pvec
;
1873 struct inode
*inode
= mpd
->inode
;
1874 struct buffer_head
*head
, *bh
;
1875 int bpp_bits
= PAGE_CACHE_SHIFT
- inode
->i_blkbits
;
1881 start
= mpd
->map
.m_lblk
>> bpp_bits
;
1882 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
1883 lblk
= start
<< bpp_bits
;
1884 pblock
= mpd
->map
.m_pblk
;
1886 pagevec_init(&pvec
, 0);
1887 while (start
<= end
) {
1888 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
1892 for (i
= 0; i
< nr_pages
; i
++) {
1893 struct page
*page
= pvec
.pages
[i
];
1895 if (page
->index
> end
)
1897 /* Up to 'end' pages must be contiguous */
1898 BUG_ON(page
->index
!= start
);
1899 bh
= head
= page_buffers(page
);
1901 if (lblk
< mpd
->map
.m_lblk
)
1903 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
1905 * Buffer after end of mapped extent.
1906 * Find next buffer in the page to map.
1909 mpd
->map
.m_flags
= 0;
1911 * FIXME: If dioread_nolock supports
1912 * blocksize < pagesize, we need to make
1913 * sure we add size mapped so far to
1914 * io_end->size as the following call
1915 * can submit the page for IO.
1917 err
= mpage_process_page_bufs(mpd
, head
,
1919 pagevec_release(&pvec
);
1924 if (buffer_delay(bh
)) {
1925 clear_buffer_delay(bh
);
1926 bh
->b_blocknr
= pblock
++;
1928 clear_buffer_unwritten(bh
);
1929 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
1932 * FIXME: This is going to break if dioread_nolock
1933 * supports blocksize < pagesize as we will try to
1934 * convert potentially unmapped parts of inode.
1936 mpd
->io_submit
.io_end
->size
+= PAGE_CACHE_SIZE
;
1937 /* Page fully mapped - let IO run! */
1938 err
= mpage_submit_page(mpd
, page
);
1940 pagevec_release(&pvec
);
1945 pagevec_release(&pvec
);
1947 /* Extent fully mapped and matches with page boundary. We are done. */
1949 mpd
->map
.m_flags
= 0;
1953 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
1955 struct inode
*inode
= mpd
->inode
;
1956 struct ext4_map_blocks
*map
= &mpd
->map
;
1957 int get_blocks_flags
;
1958 int err
, dioread_nolock
;
1960 trace_ext4_da_write_pages_extent(inode
, map
);
1962 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
1963 * to convert an unwritten extent to be initialized (in the case
1964 * where we have written into one or more preallocated blocks). It is
1965 * possible that we're going to need more metadata blocks than
1966 * previously reserved. However we must not fail because we're in
1967 * writeback and there is nothing we can do about it so it might result
1968 * in data loss. So use reserved blocks to allocate metadata if
1971 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
1972 * the blocks in question are delalloc blocks. This indicates
1973 * that the blocks and quotas has already been checked when
1974 * the data was copied into the page cache.
1976 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
1977 EXT4_GET_BLOCKS_METADATA_NOFAIL
;
1978 dioread_nolock
= ext4_should_dioread_nolock(inode
);
1980 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1981 if (map
->m_flags
& (1 << BH_Delay
))
1982 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1984 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
1987 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
1988 if (!mpd
->io_submit
.io_end
->handle
&&
1989 ext4_handle_valid(handle
)) {
1990 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
1991 handle
->h_rsv_handle
= NULL
;
1993 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
1996 BUG_ON(map
->m_len
== 0);
1997 if (map
->m_flags
& EXT4_MAP_NEW
) {
1998 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2001 for (i
= 0; i
< map
->m_len
; i
++)
2002 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2008 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2009 * mpd->len and submit pages underlying it for IO
2011 * @handle - handle for journal operations
2012 * @mpd - extent to map
2013 * @give_up_on_write - we set this to true iff there is a fatal error and there
2014 * is no hope of writing the data. The caller should discard
2015 * dirty pages to avoid infinite loops.
2017 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2018 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2019 * them to initialized or split the described range from larger unwritten
2020 * extent. Note that we need not map all the described range since allocation
2021 * can return less blocks or the range is covered by more unwritten extents. We
2022 * cannot map more because we are limited by reserved transaction credits. On
2023 * the other hand we always make sure that the last touched page is fully
2024 * mapped so that it can be written out (and thus forward progress is
2025 * guaranteed). After mapping we submit all mapped pages for IO.
2027 static int mpage_map_and_submit_extent(handle_t
*handle
,
2028 struct mpage_da_data
*mpd
,
2029 bool *give_up_on_write
)
2031 struct inode
*inode
= mpd
->inode
;
2032 struct ext4_map_blocks
*map
= &mpd
->map
;
2037 mpd
->io_submit
.io_end
->offset
=
2038 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2040 err
= mpage_map_one_extent(handle
, mpd
);
2042 struct super_block
*sb
= inode
->i_sb
;
2044 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2045 goto invalidate_dirty_pages
;
2047 * Let the uper layers retry transient errors.
2048 * In the case of ENOSPC, if ext4_count_free_blocks()
2049 * is non-zero, a commit should free up blocks.
2051 if ((err
== -ENOMEM
) ||
2052 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2054 goto update_disksize
;
2057 ext4_msg(sb
, KERN_CRIT
,
2058 "Delayed block allocation failed for "
2059 "inode %lu at logical offset %llu with"
2060 " max blocks %u with error %d",
2062 (unsigned long long)map
->m_lblk
,
2063 (unsigned)map
->m_len
, -err
);
2064 ext4_msg(sb
, KERN_CRIT
,
2065 "This should not happen!! Data will "
2068 ext4_print_free_blocks(inode
);
2069 invalidate_dirty_pages
:
2070 *give_up_on_write
= true;
2075 * Update buffer state, submit mapped pages, and get us new
2078 err
= mpage_map_and_submit_buffers(mpd
);
2080 goto update_disksize
;
2081 } while (map
->m_len
);
2085 * Update on-disk size after IO is submitted. Races with
2086 * truncate are avoided by checking i_size under i_data_sem.
2088 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_CACHE_SHIFT
;
2089 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2093 down_write(&EXT4_I(inode
)->i_data_sem
);
2094 i_size
= i_size_read(inode
);
2095 if (disksize
> i_size
)
2097 if (disksize
> EXT4_I(inode
)->i_disksize
)
2098 EXT4_I(inode
)->i_disksize
= disksize
;
2099 err2
= ext4_mark_inode_dirty(handle
, inode
);
2100 up_write(&EXT4_I(inode
)->i_data_sem
);
2102 ext4_error(inode
->i_sb
,
2103 "Failed to mark inode %lu dirty",
2112 * Calculate the total number of credits to reserve for one writepages
2113 * iteration. This is called from ext4_writepages(). We map an extent of
2114 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2115 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2116 * bpp - 1 blocks in bpp different extents.
2118 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2120 int bpp
= ext4_journal_blocks_per_page(inode
);
2122 return ext4_meta_trans_blocks(inode
,
2123 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2127 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2128 * and underlying extent to map
2130 * @mpd - where to look for pages
2132 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2133 * IO immediately. When we find a page which isn't mapped we start accumulating
2134 * extent of buffers underlying these pages that needs mapping (formed by
2135 * either delayed or unwritten buffers). We also lock the pages containing
2136 * these buffers. The extent found is returned in @mpd structure (starting at
2137 * mpd->lblk with length mpd->len blocks).
2139 * Note that this function can attach bios to one io_end structure which are
2140 * neither logically nor physically contiguous. Although it may seem as an
2141 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2142 * case as we need to track IO to all buffers underlying a page in one io_end.
2144 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2146 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2147 struct pagevec pvec
;
2148 unsigned int nr_pages
;
2149 long left
= mpd
->wbc
->nr_to_write
;
2150 pgoff_t index
= mpd
->first_page
;
2151 pgoff_t end
= mpd
->last_page
;
2154 int blkbits
= mpd
->inode
->i_blkbits
;
2156 struct buffer_head
*head
;
2158 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2159 tag
= PAGECACHE_TAG_TOWRITE
;
2161 tag
= PAGECACHE_TAG_DIRTY
;
2163 pagevec_init(&pvec
, 0);
2165 mpd
->next_page
= index
;
2166 while (index
<= end
) {
2167 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2168 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2172 for (i
= 0; i
< nr_pages
; i
++) {
2173 struct page
*page
= pvec
.pages
[i
];
2176 * At this point, the page may be truncated or
2177 * invalidated (changing page->mapping to NULL), or
2178 * even swizzled back from swapper_space to tmpfs file
2179 * mapping. However, page->index will not change
2180 * because we have a reference on the page.
2182 if (page
->index
> end
)
2186 * Accumulated enough dirty pages? This doesn't apply
2187 * to WB_SYNC_ALL mode. For integrity sync we have to
2188 * keep going because someone may be concurrently
2189 * dirtying pages, and we might have synced a lot of
2190 * newly appeared dirty pages, but have not synced all
2191 * of the old dirty pages.
2193 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2196 /* If we can't merge this page, we are done. */
2197 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2202 * If the page is no longer dirty, or its mapping no
2203 * longer corresponds to inode we are writing (which
2204 * means it has been truncated or invalidated), or the
2205 * page is already under writeback and we are not doing
2206 * a data integrity writeback, skip the page
2208 if (!PageDirty(page
) ||
2209 (PageWriteback(page
) &&
2210 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2211 unlikely(page
->mapping
!= mapping
)) {
2216 wait_on_page_writeback(page
);
2217 BUG_ON(PageWriteback(page
));
2219 if (mpd
->map
.m_len
== 0)
2220 mpd
->first_page
= page
->index
;
2221 mpd
->next_page
= page
->index
+ 1;
2222 /* Add all dirty buffers to mpd */
2223 lblk
= ((ext4_lblk_t
)page
->index
) <<
2224 (PAGE_CACHE_SHIFT
- blkbits
);
2225 head
= page_buffers(page
);
2226 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2232 pagevec_release(&pvec
);
2237 pagevec_release(&pvec
);
2241 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2244 struct address_space
*mapping
= data
;
2245 int ret
= ext4_writepage(page
, wbc
);
2246 mapping_set_error(mapping
, ret
);
2250 static int ext4_writepages(struct address_space
*mapping
,
2251 struct writeback_control
*wbc
)
2253 pgoff_t writeback_index
= 0;
2254 long nr_to_write
= wbc
->nr_to_write
;
2255 int range_whole
= 0;
2257 handle_t
*handle
= NULL
;
2258 struct mpage_da_data mpd
;
2259 struct inode
*inode
= mapping
->host
;
2260 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2261 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2263 struct blk_plug plug
;
2264 bool give_up_on_write
= false;
2266 trace_ext4_writepages(inode
, wbc
);
2269 * No pages to write? This is mainly a kludge to avoid starting
2270 * a transaction for special inodes like journal inode on last iput()
2271 * because that could violate lock ordering on umount
2273 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2274 goto out_writepages
;
2276 if (ext4_should_journal_data(inode
)) {
2277 struct blk_plug plug
;
2279 blk_start_plug(&plug
);
2280 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2281 blk_finish_plug(&plug
);
2282 goto out_writepages
;
2286 * If the filesystem has aborted, it is read-only, so return
2287 * right away instead of dumping stack traces later on that
2288 * will obscure the real source of the problem. We test
2289 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2290 * the latter could be true if the filesystem is mounted
2291 * read-only, and in that case, ext4_writepages should
2292 * *never* be called, so if that ever happens, we would want
2295 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2297 goto out_writepages
;
2300 if (ext4_should_dioread_nolock(inode
)) {
2302 * We may need to convert up to one extent per block in
2303 * the page and we may dirty the inode.
2305 rsv_blocks
= 1 + (PAGE_CACHE_SIZE
>> inode
->i_blkbits
);
2309 * If we have inline data and arrive here, it means that
2310 * we will soon create the block for the 1st page, so
2311 * we'd better clear the inline data here.
2313 if (ext4_has_inline_data(inode
)) {
2314 /* Just inode will be modified... */
2315 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2316 if (IS_ERR(handle
)) {
2317 ret
= PTR_ERR(handle
);
2318 goto out_writepages
;
2320 BUG_ON(ext4_test_inode_state(inode
,
2321 EXT4_STATE_MAY_INLINE_DATA
));
2322 ext4_destroy_inline_data(handle
, inode
);
2323 ext4_journal_stop(handle
);
2326 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2329 if (wbc
->range_cyclic
) {
2330 writeback_index
= mapping
->writeback_index
;
2331 if (writeback_index
)
2333 mpd
.first_page
= writeback_index
;
2336 mpd
.first_page
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2337 mpd
.last_page
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2342 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2344 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2345 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2347 blk_start_plug(&plug
);
2348 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2349 /* For each extent of pages we use new io_end */
2350 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2351 if (!mpd
.io_submit
.io_end
) {
2357 * We have two constraints: We find one extent to map and we
2358 * must always write out whole page (makes a difference when
2359 * blocksize < pagesize) so that we don't block on IO when we
2360 * try to write out the rest of the page. Journalled mode is
2361 * not supported by delalloc.
2363 BUG_ON(ext4_should_journal_data(inode
));
2364 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2366 /* start a new transaction */
2367 handle
= ext4_journal_start_with_reserve(inode
,
2368 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2369 if (IS_ERR(handle
)) {
2370 ret
= PTR_ERR(handle
);
2371 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2372 "%ld pages, ino %lu; err %d", __func__
,
2373 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2374 /* Release allocated io_end */
2375 ext4_put_io_end(mpd
.io_submit
.io_end
);
2379 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2380 ret
= mpage_prepare_extent_to_map(&mpd
);
2383 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2387 * We scanned the whole range (or exhausted
2388 * nr_to_write), submitted what was mapped and
2389 * didn't find anything needing mapping. We are
2395 ext4_journal_stop(handle
);
2396 /* Submit prepared bio */
2397 ext4_io_submit(&mpd
.io_submit
);
2398 /* Unlock pages we didn't use */
2399 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2400 /* Drop our io_end reference we got from init */
2401 ext4_put_io_end(mpd
.io_submit
.io_end
);
2403 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2405 * Commit the transaction which would
2406 * free blocks released in the transaction
2409 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2413 /* Fatal error - ENOMEM, EIO... */
2417 blk_finish_plug(&plug
);
2418 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2420 mpd
.last_page
= writeback_index
- 1;
2426 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2428 * Set the writeback_index so that range_cyclic
2429 * mode will write it back later
2431 mapping
->writeback_index
= mpd
.first_page
;
2434 trace_ext4_writepages_result(inode
, wbc
, ret
,
2435 nr_to_write
- wbc
->nr_to_write
);
2439 static int ext4_nonda_switch(struct super_block
*sb
)
2441 s64 free_clusters
, dirty_clusters
;
2442 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2445 * switch to non delalloc mode if we are running low
2446 * on free block. The free block accounting via percpu
2447 * counters can get slightly wrong with percpu_counter_batch getting
2448 * accumulated on each CPU without updating global counters
2449 * Delalloc need an accurate free block accounting. So switch
2450 * to non delalloc when we are near to error range.
2453 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2455 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2457 * Start pushing delalloc when 1/2 of free blocks are dirty.
2459 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2460 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2462 if (2 * free_clusters
< 3 * dirty_clusters
||
2463 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2465 * free block count is less than 150% of dirty blocks
2466 * or free blocks is less than watermark
2473 /* We always reserve for an inode update; the superblock could be there too */
2474 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2476 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
2477 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
)))
2480 if (pos
+ len
<= 0x7fffffffULL
)
2483 /* We might need to update the superblock to set LARGE_FILE */
2487 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2488 loff_t pos
, unsigned len
, unsigned flags
,
2489 struct page
**pagep
, void **fsdata
)
2491 int ret
, retries
= 0;
2494 struct inode
*inode
= mapping
->host
;
2497 index
= pos
>> PAGE_CACHE_SHIFT
;
2499 if (ext4_nonda_switch(inode
->i_sb
)) {
2500 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2501 return ext4_write_begin(file
, mapping
, pos
,
2502 len
, flags
, pagep
, fsdata
);
2504 *fsdata
= (void *)0;
2505 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2507 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2508 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2518 * grab_cache_page_write_begin() can take a long time if the
2519 * system is thrashing due to memory pressure, or if the page
2520 * is being written back. So grab it first before we start
2521 * the transaction handle. This also allows us to allocate
2522 * the page (if needed) without using GFP_NOFS.
2525 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2531 * With delayed allocation, we don't log the i_disksize update
2532 * if there is delayed block allocation. But we still need
2533 * to journalling the i_disksize update if writes to the end
2534 * of file which has an already mapped buffer.
2537 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2538 ext4_da_write_credits(inode
, pos
, len
));
2539 if (IS_ERR(handle
)) {
2540 page_cache_release(page
);
2541 return PTR_ERR(handle
);
2545 if (page
->mapping
!= mapping
) {
2546 /* The page got truncated from under us */
2548 page_cache_release(page
);
2549 ext4_journal_stop(handle
);
2552 /* In case writeback began while the page was unlocked */
2553 wait_for_stable_page(page
);
2555 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2558 ext4_journal_stop(handle
);
2560 * block_write_begin may have instantiated a few blocks
2561 * outside i_size. Trim these off again. Don't need
2562 * i_size_read because we hold i_mutex.
2564 if (pos
+ len
> inode
->i_size
)
2565 ext4_truncate_failed_write(inode
);
2567 if (ret
== -ENOSPC
&&
2568 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2571 page_cache_release(page
);
2580 * Check if we should update i_disksize
2581 * when write to the end of file but not require block allocation
2583 static int ext4_da_should_update_i_disksize(struct page
*page
,
2584 unsigned long offset
)
2586 struct buffer_head
*bh
;
2587 struct inode
*inode
= page
->mapping
->host
;
2591 bh
= page_buffers(page
);
2592 idx
= offset
>> inode
->i_blkbits
;
2594 for (i
= 0; i
< idx
; i
++)
2595 bh
= bh
->b_this_page
;
2597 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2602 static int ext4_da_write_end(struct file
*file
,
2603 struct address_space
*mapping
,
2604 loff_t pos
, unsigned len
, unsigned copied
,
2605 struct page
*page
, void *fsdata
)
2607 struct inode
*inode
= mapping
->host
;
2609 handle_t
*handle
= ext4_journal_current_handle();
2611 unsigned long start
, end
;
2612 int write_mode
= (int)(unsigned long)fsdata
;
2614 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2615 return ext4_write_end(file
, mapping
, pos
,
2616 len
, copied
, page
, fsdata
);
2618 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2619 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2620 end
= start
+ copied
- 1;
2623 * generic_write_end() will run mark_inode_dirty() if i_size
2624 * changes. So let's piggyback the i_disksize mark_inode_dirty
2627 new_i_size
= pos
+ copied
;
2628 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2629 if (ext4_has_inline_data(inode
) ||
2630 ext4_da_should_update_i_disksize(page
, end
)) {
2631 ext4_update_i_disksize(inode
, new_i_size
);
2632 /* We need to mark inode dirty even if
2633 * new_i_size is less that inode->i_size
2634 * bu greater than i_disksize.(hint delalloc)
2636 ext4_mark_inode_dirty(handle
, inode
);
2640 if (write_mode
!= CONVERT_INLINE_DATA
&&
2641 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2642 ext4_has_inline_data(inode
))
2643 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2646 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2652 ret2
= ext4_journal_stop(handle
);
2656 return ret
? ret
: copied
;
2659 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
2660 unsigned int length
)
2663 * Drop reserved blocks
2665 BUG_ON(!PageLocked(page
));
2666 if (!page_has_buffers(page
))
2669 ext4_da_page_release_reservation(page
, offset
, length
);
2672 ext4_invalidatepage(page
, offset
, length
);
2678 * Force all delayed allocation blocks to be allocated for a given inode.
2680 int ext4_alloc_da_blocks(struct inode
*inode
)
2682 trace_ext4_alloc_da_blocks(inode
);
2684 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
2688 * We do something simple for now. The filemap_flush() will
2689 * also start triggering a write of the data blocks, which is
2690 * not strictly speaking necessary (and for users of
2691 * laptop_mode, not even desirable). However, to do otherwise
2692 * would require replicating code paths in:
2694 * ext4_writepages() ->
2695 * write_cache_pages() ---> (via passed in callback function)
2696 * __mpage_da_writepage() -->
2697 * mpage_add_bh_to_extent()
2698 * mpage_da_map_blocks()
2700 * The problem is that write_cache_pages(), located in
2701 * mm/page-writeback.c, marks pages clean in preparation for
2702 * doing I/O, which is not desirable if we're not planning on
2705 * We could call write_cache_pages(), and then redirty all of
2706 * the pages by calling redirty_page_for_writepage() but that
2707 * would be ugly in the extreme. So instead we would need to
2708 * replicate parts of the code in the above functions,
2709 * simplifying them because we wouldn't actually intend to
2710 * write out the pages, but rather only collect contiguous
2711 * logical block extents, call the multi-block allocator, and
2712 * then update the buffer heads with the block allocations.
2714 * For now, though, we'll cheat by calling filemap_flush(),
2715 * which will map the blocks, and start the I/O, but not
2716 * actually wait for the I/O to complete.
2718 return filemap_flush(inode
->i_mapping
);
2722 * bmap() is special. It gets used by applications such as lilo and by
2723 * the swapper to find the on-disk block of a specific piece of data.
2725 * Naturally, this is dangerous if the block concerned is still in the
2726 * journal. If somebody makes a swapfile on an ext4 data-journaling
2727 * filesystem and enables swap, then they may get a nasty shock when the
2728 * data getting swapped to that swapfile suddenly gets overwritten by
2729 * the original zero's written out previously to the journal and
2730 * awaiting writeback in the kernel's buffer cache.
2732 * So, if we see any bmap calls here on a modified, data-journaled file,
2733 * take extra steps to flush any blocks which might be in the cache.
2735 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2737 struct inode
*inode
= mapping
->host
;
2742 * We can get here for an inline file via the FIBMAP ioctl
2744 if (ext4_has_inline_data(inode
))
2747 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2748 test_opt(inode
->i_sb
, DELALLOC
)) {
2750 * With delalloc we want to sync the file
2751 * so that we can make sure we allocate
2754 filemap_write_and_wait(mapping
);
2757 if (EXT4_JOURNAL(inode
) &&
2758 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2760 * This is a REALLY heavyweight approach, but the use of
2761 * bmap on dirty files is expected to be extremely rare:
2762 * only if we run lilo or swapon on a freshly made file
2763 * do we expect this to happen.
2765 * (bmap requires CAP_SYS_RAWIO so this does not
2766 * represent an unprivileged user DOS attack --- we'd be
2767 * in trouble if mortal users could trigger this path at
2770 * NB. EXT4_STATE_JDATA is not set on files other than
2771 * regular files. If somebody wants to bmap a directory
2772 * or symlink and gets confused because the buffer
2773 * hasn't yet been flushed to disk, they deserve
2774 * everything they get.
2777 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2778 journal
= EXT4_JOURNAL(inode
);
2779 jbd2_journal_lock_updates(journal
);
2780 err
= jbd2_journal_flush(journal
);
2781 jbd2_journal_unlock_updates(journal
);
2787 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2790 static int ext4_readpage(struct file
*file
, struct page
*page
)
2793 struct inode
*inode
= page
->mapping
->host
;
2795 trace_ext4_readpage(page
);
2797 if (ext4_has_inline_data(inode
))
2798 ret
= ext4_readpage_inline(inode
, page
);
2801 return mpage_readpage(page
, ext4_get_block
);
2807 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2808 struct list_head
*pages
, unsigned nr_pages
)
2810 struct inode
*inode
= mapping
->host
;
2812 /* If the file has inline data, no need to do readpages. */
2813 if (ext4_has_inline_data(inode
))
2816 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2819 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
2820 unsigned int length
)
2822 trace_ext4_invalidatepage(page
, offset
, length
);
2824 /* No journalling happens on data buffers when this function is used */
2825 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
2827 block_invalidatepage(page
, offset
, length
);
2830 static int __ext4_journalled_invalidatepage(struct page
*page
,
2831 unsigned int offset
,
2832 unsigned int length
)
2834 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2836 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
2839 * If it's a full truncate we just forget about the pending dirtying
2841 if (offset
== 0 && length
== PAGE_CACHE_SIZE
)
2842 ClearPageChecked(page
);
2844 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
2847 /* Wrapper for aops... */
2848 static void ext4_journalled_invalidatepage(struct page
*page
,
2849 unsigned int offset
,
2850 unsigned int length
)
2852 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
2855 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2857 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2859 trace_ext4_releasepage(page
);
2861 /* Page has dirty journalled data -> cannot release */
2862 if (PageChecked(page
))
2865 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2867 return try_to_free_buffers(page
);
2871 * ext4_get_block used when preparing for a DIO write or buffer write.
2872 * We allocate an uinitialized extent if blocks haven't been allocated.
2873 * The extent will be converted to initialized after the IO is complete.
2875 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
2876 struct buffer_head
*bh_result
, int create
)
2878 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2879 inode
->i_ino
, create
);
2880 return _ext4_get_block(inode
, iblock
, bh_result
,
2881 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
2884 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
2885 struct buffer_head
*bh_result
, int create
)
2887 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2888 inode
->i_ino
, create
);
2889 return _ext4_get_block(inode
, iblock
, bh_result
,
2890 EXT4_GET_BLOCKS_NO_LOCK
);
2893 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
2894 ssize_t size
, void *private)
2896 ext4_io_end_t
*io_end
= iocb
->private;
2898 /* if not async direct IO just return */
2902 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2903 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2904 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
2907 iocb
->private = NULL
;
2908 io_end
->offset
= offset
;
2909 io_end
->size
= size
;
2910 ext4_put_io_end(io_end
);
2914 * For ext4 extent files, ext4 will do direct-io write to holes,
2915 * preallocated extents, and those write extend the file, no need to
2916 * fall back to buffered IO.
2918 * For holes, we fallocate those blocks, mark them as unwritten
2919 * If those blocks were preallocated, we mark sure they are split, but
2920 * still keep the range to write as unwritten.
2922 * The unwritten extents will be converted to written when DIO is completed.
2923 * For async direct IO, since the IO may still pending when return, we
2924 * set up an end_io call back function, which will do the conversion
2925 * when async direct IO completed.
2927 * If the O_DIRECT write will extend the file then add this inode to the
2928 * orphan list. So recovery will truncate it back to the original size
2929 * if the machine crashes during the write.
2932 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
2933 struct iov_iter
*iter
, loff_t offset
)
2935 struct file
*file
= iocb
->ki_filp
;
2936 struct inode
*inode
= file
->f_mapping
->host
;
2938 size_t count
= iov_iter_count(iter
);
2940 get_block_t
*get_block_func
= NULL
;
2942 loff_t final_size
= offset
+ count
;
2943 ext4_io_end_t
*io_end
= NULL
;
2945 /* Use the old path for reads and writes beyond i_size. */
2946 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
2947 return ext4_ind_direct_IO(rw
, iocb
, iter
, offset
);
2949 BUG_ON(iocb
->private == NULL
);
2952 * Make all waiters for direct IO properly wait also for extent
2953 * conversion. This also disallows race between truncate() and
2954 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
2957 atomic_inc(&inode
->i_dio_count
);
2959 /* If we do a overwrite dio, i_mutex locking can be released */
2960 overwrite
= *((int *)iocb
->private);
2963 down_read(&EXT4_I(inode
)->i_data_sem
);
2964 mutex_unlock(&inode
->i_mutex
);
2968 * We could direct write to holes and fallocate.
2970 * Allocated blocks to fill the hole are marked as
2971 * unwritten to prevent parallel buffered read to expose
2972 * the stale data before DIO complete the data IO.
2974 * As to previously fallocated extents, ext4 get_block will
2975 * just simply mark the buffer mapped but still keep the
2976 * extents unwritten.
2978 * For non AIO case, we will convert those unwritten extents
2979 * to written after return back from blockdev_direct_IO.
2981 * For async DIO, the conversion needs to be deferred when the
2982 * IO is completed. The ext4 end_io callback function will be
2983 * called to take care of the conversion work. Here for async
2984 * case, we allocate an io_end structure to hook to the iocb.
2986 iocb
->private = NULL
;
2987 ext4_inode_aio_set(inode
, NULL
);
2988 if (!is_sync_kiocb(iocb
)) {
2989 io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2995 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
2997 iocb
->private = ext4_get_io_end(io_end
);
2999 * we save the io structure for current async direct
3000 * IO, so that later ext4_map_blocks() could flag the
3001 * io structure whether there is a unwritten extents
3002 * needs to be converted when IO is completed.
3004 ext4_inode_aio_set(inode
, io_end
);
3008 get_block_func
= ext4_get_block_write_nolock
;
3010 get_block_func
= ext4_get_block_write
;
3011 dio_flags
= DIO_LOCKING
;
3013 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3014 inode
->i_sb
->s_bdev
, iter
,
3022 * Put our reference to io_end. This can free the io_end structure e.g.
3023 * in sync IO case or in case of error. It can even perform extent
3024 * conversion if all bios we submitted finished before we got here.
3025 * Note that in that case iocb->private can be already set to NULL
3029 ext4_inode_aio_set(inode
, NULL
);
3030 ext4_put_io_end(io_end
);
3032 * When no IO was submitted ext4_end_io_dio() was not
3033 * called so we have to put iocb's reference.
3035 if (ret
<= 0 && ret
!= -EIOCBQUEUED
&& iocb
->private) {
3036 WARN_ON(iocb
->private != io_end
);
3037 WARN_ON(io_end
->flag
& EXT4_IO_END_UNWRITTEN
);
3038 ext4_put_io_end(io_end
);
3039 iocb
->private = NULL
;
3042 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3043 EXT4_STATE_DIO_UNWRITTEN
)) {
3046 * for non AIO case, since the IO is already
3047 * completed, we could do the conversion right here
3049 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3053 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3058 inode_dio_done(inode
);
3059 /* take i_mutex locking again if we do a ovewrite dio */
3061 up_read(&EXT4_I(inode
)->i_data_sem
);
3062 mutex_lock(&inode
->i_mutex
);
3068 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3069 struct iov_iter
*iter
, loff_t offset
)
3071 struct file
*file
= iocb
->ki_filp
;
3072 struct inode
*inode
= file
->f_mapping
->host
;
3073 size_t count
= iov_iter_count(iter
);
3077 * If we are doing data journalling we don't support O_DIRECT
3079 if (ext4_should_journal_data(inode
))
3082 /* Let buffer I/O handle the inline data case. */
3083 if (ext4_has_inline_data(inode
))
3086 trace_ext4_direct_IO_enter(inode
, offset
, count
, rw
);
3087 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3088 ret
= ext4_ext_direct_IO(rw
, iocb
, iter
, offset
);
3090 ret
= ext4_ind_direct_IO(rw
, iocb
, iter
, offset
);
3091 trace_ext4_direct_IO_exit(inode
, offset
, count
, rw
, ret
);
3096 * Pages can be marked dirty completely asynchronously from ext4's journalling
3097 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3098 * much here because ->set_page_dirty is called under VFS locks. The page is
3099 * not necessarily locked.
3101 * We cannot just dirty the page and leave attached buffers clean, because the
3102 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3103 * or jbddirty because all the journalling code will explode.
3105 * So what we do is to mark the page "pending dirty" and next time writepage
3106 * is called, propagate that into the buffers appropriately.
3108 static int ext4_journalled_set_page_dirty(struct page
*page
)
3110 SetPageChecked(page
);
3111 return __set_page_dirty_nobuffers(page
);
3114 static const struct address_space_operations ext4_aops
= {
3115 .readpage
= ext4_readpage
,
3116 .readpages
= ext4_readpages
,
3117 .writepage
= ext4_writepage
,
3118 .writepages
= ext4_writepages
,
3119 .write_begin
= ext4_write_begin
,
3120 .write_end
= ext4_write_end
,
3122 .invalidatepage
= ext4_invalidatepage
,
3123 .releasepage
= ext4_releasepage
,
3124 .direct_IO
= ext4_direct_IO
,
3125 .migratepage
= buffer_migrate_page
,
3126 .is_partially_uptodate
= block_is_partially_uptodate
,
3127 .error_remove_page
= generic_error_remove_page
,
3130 static const struct address_space_operations ext4_journalled_aops
= {
3131 .readpage
= ext4_readpage
,
3132 .readpages
= ext4_readpages
,
3133 .writepage
= ext4_writepage
,
3134 .writepages
= ext4_writepages
,
3135 .write_begin
= ext4_write_begin
,
3136 .write_end
= ext4_journalled_write_end
,
3137 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3139 .invalidatepage
= ext4_journalled_invalidatepage
,
3140 .releasepage
= ext4_releasepage
,
3141 .direct_IO
= ext4_direct_IO
,
3142 .is_partially_uptodate
= block_is_partially_uptodate
,
3143 .error_remove_page
= generic_error_remove_page
,
3146 static const struct address_space_operations ext4_da_aops
= {
3147 .readpage
= ext4_readpage
,
3148 .readpages
= ext4_readpages
,
3149 .writepage
= ext4_writepage
,
3150 .writepages
= ext4_writepages
,
3151 .write_begin
= ext4_da_write_begin
,
3152 .write_end
= ext4_da_write_end
,
3154 .invalidatepage
= ext4_da_invalidatepage
,
3155 .releasepage
= ext4_releasepage
,
3156 .direct_IO
= ext4_direct_IO
,
3157 .migratepage
= buffer_migrate_page
,
3158 .is_partially_uptodate
= block_is_partially_uptodate
,
3159 .error_remove_page
= generic_error_remove_page
,
3162 void ext4_set_aops(struct inode
*inode
)
3164 switch (ext4_inode_journal_mode(inode
)) {
3165 case EXT4_INODE_ORDERED_DATA_MODE
:
3166 ext4_set_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3168 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3169 ext4_clear_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3171 case EXT4_INODE_JOURNAL_DATA_MODE
:
3172 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3177 if (test_opt(inode
->i_sb
, DELALLOC
))
3178 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3180 inode
->i_mapping
->a_ops
= &ext4_aops
;
3184 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3185 * starting from file offset 'from'. The range to be zero'd must
3186 * be contained with in one block. If the specified range exceeds
3187 * the end of the block it will be shortened to end of the block
3188 * that cooresponds to 'from'
3190 static int ext4_block_zero_page_range(handle_t
*handle
,
3191 struct address_space
*mapping
, loff_t from
, loff_t length
)
3193 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3194 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3195 unsigned blocksize
, max
, pos
;
3197 struct inode
*inode
= mapping
->host
;
3198 struct buffer_head
*bh
;
3202 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3203 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3207 blocksize
= inode
->i_sb
->s_blocksize
;
3208 max
= blocksize
- (offset
& (blocksize
- 1));
3211 * correct length if it does not fall between
3212 * 'from' and the end of the block
3214 if (length
> max
|| length
< 0)
3217 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3219 if (!page_has_buffers(page
))
3220 create_empty_buffers(page
, blocksize
, 0);
3222 /* Find the buffer that contains "offset" */
3223 bh
= page_buffers(page
);
3225 while (offset
>= pos
) {
3226 bh
= bh
->b_this_page
;
3230 if (buffer_freed(bh
)) {
3231 BUFFER_TRACE(bh
, "freed: skip");
3234 if (!buffer_mapped(bh
)) {
3235 BUFFER_TRACE(bh
, "unmapped");
3236 ext4_get_block(inode
, iblock
, bh
, 0);
3237 /* unmapped? It's a hole - nothing to do */
3238 if (!buffer_mapped(bh
)) {
3239 BUFFER_TRACE(bh
, "still unmapped");
3244 /* Ok, it's mapped. Make sure it's up-to-date */
3245 if (PageUptodate(page
))
3246 set_buffer_uptodate(bh
);
3248 if (!buffer_uptodate(bh
)) {
3250 ll_rw_block(READ
, 1, &bh
);
3252 /* Uhhuh. Read error. Complain and punt. */
3253 if (!buffer_uptodate(bh
))
3256 if (ext4_should_journal_data(inode
)) {
3257 BUFFER_TRACE(bh
, "get write access");
3258 err
= ext4_journal_get_write_access(handle
, bh
);
3262 zero_user(page
, offset
, length
);
3263 BUFFER_TRACE(bh
, "zeroed end of block");
3265 if (ext4_should_journal_data(inode
)) {
3266 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3269 mark_buffer_dirty(bh
);
3270 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
))
3271 err
= ext4_jbd2_file_inode(handle
, inode
);
3276 page_cache_release(page
);
3281 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3282 * up to the end of the block which corresponds to `from'.
3283 * This required during truncate. We need to physically zero the tail end
3284 * of that block so it doesn't yield old data if the file is later grown.
3286 static int ext4_block_truncate_page(handle_t
*handle
,
3287 struct address_space
*mapping
, loff_t from
)
3289 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3292 struct inode
*inode
= mapping
->host
;
3294 blocksize
= inode
->i_sb
->s_blocksize
;
3295 length
= blocksize
- (offset
& (blocksize
- 1));
3297 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3300 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3301 loff_t lstart
, loff_t length
)
3303 struct super_block
*sb
= inode
->i_sb
;
3304 struct address_space
*mapping
= inode
->i_mapping
;
3305 unsigned partial_start
, partial_end
;
3306 ext4_fsblk_t start
, end
;
3307 loff_t byte_end
= (lstart
+ length
- 1);
3310 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3311 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3313 start
= lstart
>> sb
->s_blocksize_bits
;
3314 end
= byte_end
>> sb
->s_blocksize_bits
;
3316 /* Handle partial zero within the single block */
3318 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3319 err
= ext4_block_zero_page_range(handle
, mapping
,
3323 /* Handle partial zero out on the start of the range */
3324 if (partial_start
) {
3325 err
= ext4_block_zero_page_range(handle
, mapping
,
3326 lstart
, sb
->s_blocksize
);
3330 /* Handle partial zero out on the end of the range */
3331 if (partial_end
!= sb
->s_blocksize
- 1)
3332 err
= ext4_block_zero_page_range(handle
, mapping
,
3333 byte_end
- partial_end
,
3338 int ext4_can_truncate(struct inode
*inode
)
3340 if (S_ISREG(inode
->i_mode
))
3342 if (S_ISDIR(inode
->i_mode
))
3344 if (S_ISLNK(inode
->i_mode
))
3345 return !ext4_inode_is_fast_symlink(inode
);
3350 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3351 * associated with the given offset and length
3353 * @inode: File inode
3354 * @offset: The offset where the hole will begin
3355 * @len: The length of the hole
3357 * Returns: 0 on success or negative on failure
3360 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3362 struct super_block
*sb
= inode
->i_sb
;
3363 ext4_lblk_t first_block
, stop_block
;
3364 struct address_space
*mapping
= inode
->i_mapping
;
3365 loff_t first_block_offset
, last_block_offset
;
3367 unsigned int credits
;
3370 if (!S_ISREG(inode
->i_mode
))
3373 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3376 * Write out all dirty pages to avoid race conditions
3377 * Then release them.
3379 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3380 ret
= filemap_write_and_wait_range(mapping
, offset
,
3381 offset
+ length
- 1);
3386 mutex_lock(&inode
->i_mutex
);
3388 /* No need to punch hole beyond i_size */
3389 if (offset
>= inode
->i_size
)
3393 * If the hole extends beyond i_size, set the hole
3394 * to end after the page that contains i_size
3396 if (offset
+ length
> inode
->i_size
) {
3397 length
= inode
->i_size
+
3398 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
3402 if (offset
& (sb
->s_blocksize
- 1) ||
3403 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
3405 * Attach jinode to inode for jbd2 if we do any zeroing of
3408 ret
= ext4_inode_attach_jinode(inode
);
3414 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
3415 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
3417 /* Now release the pages and zero block aligned part of pages*/
3418 if (last_block_offset
> first_block_offset
)
3419 truncate_pagecache_range(inode
, first_block_offset
,
3422 /* Wait all existing dio workers, newcomers will block on i_mutex */
3423 ext4_inode_block_unlocked_dio(inode
);
3424 inode_dio_wait(inode
);
3426 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3427 credits
= ext4_writepage_trans_blocks(inode
);
3429 credits
= ext4_blocks_for_truncate(inode
);
3430 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3431 if (IS_ERR(handle
)) {
3432 ret
= PTR_ERR(handle
);
3433 ext4_std_error(sb
, ret
);
3437 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
3442 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3443 EXT4_BLOCK_SIZE_BITS(sb
);
3444 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3446 /* If there are no blocks to remove, return now */
3447 if (first_block
>= stop_block
)
3450 down_write(&EXT4_I(inode
)->i_data_sem
);
3451 ext4_discard_preallocations(inode
);
3453 ret
= ext4_es_remove_extent(inode
, first_block
,
3454 stop_block
- first_block
);
3456 up_write(&EXT4_I(inode
)->i_data_sem
);
3460 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3461 ret
= ext4_ext_remove_space(inode
, first_block
,
3464 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
3467 up_write(&EXT4_I(inode
)->i_data_sem
);
3469 ext4_handle_sync(handle
);
3471 /* Now release the pages again to reduce race window */
3472 if (last_block_offset
> first_block_offset
)
3473 truncate_pagecache_range(inode
, first_block_offset
,
3476 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3477 ext4_mark_inode_dirty(handle
, inode
);
3479 ext4_journal_stop(handle
);
3481 ext4_inode_resume_unlocked_dio(inode
);
3483 mutex_unlock(&inode
->i_mutex
);
3487 int ext4_inode_attach_jinode(struct inode
*inode
)
3489 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3490 struct jbd2_inode
*jinode
;
3492 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
3495 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
3496 spin_lock(&inode
->i_lock
);
3499 spin_unlock(&inode
->i_lock
);
3502 ei
->jinode
= jinode
;
3503 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
3506 spin_unlock(&inode
->i_lock
);
3507 if (unlikely(jinode
!= NULL
))
3508 jbd2_free_inode(jinode
);
3515 * We block out ext4_get_block() block instantiations across the entire
3516 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3517 * simultaneously on behalf of the same inode.
3519 * As we work through the truncate and commit bits of it to the journal there
3520 * is one core, guiding principle: the file's tree must always be consistent on
3521 * disk. We must be able to restart the truncate after a crash.
3523 * The file's tree may be transiently inconsistent in memory (although it
3524 * probably isn't), but whenever we close off and commit a journal transaction,
3525 * the contents of (the filesystem + the journal) must be consistent and
3526 * restartable. It's pretty simple, really: bottom up, right to left (although
3527 * left-to-right works OK too).
3529 * Note that at recovery time, journal replay occurs *before* the restart of
3530 * truncate against the orphan inode list.
3532 * The committed inode has the new, desired i_size (which is the same as
3533 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3534 * that this inode's truncate did not complete and it will again call
3535 * ext4_truncate() to have another go. So there will be instantiated blocks
3536 * to the right of the truncation point in a crashed ext4 filesystem. But
3537 * that's fine - as long as they are linked from the inode, the post-crash
3538 * ext4_truncate() run will find them and release them.
3540 void ext4_truncate(struct inode
*inode
)
3542 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3543 unsigned int credits
;
3545 struct address_space
*mapping
= inode
->i_mapping
;
3548 * There is a possibility that we're either freeing the inode
3549 * or it's a completely new inode. In those cases we might not
3550 * have i_mutex locked because it's not necessary.
3552 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
3553 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
3554 trace_ext4_truncate_enter(inode
);
3556 if (!ext4_can_truncate(inode
))
3559 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3561 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3562 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3564 if (ext4_has_inline_data(inode
)) {
3567 ext4_inline_data_truncate(inode
, &has_inline
);
3572 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3573 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
3574 if (ext4_inode_attach_jinode(inode
) < 0)
3578 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3579 credits
= ext4_writepage_trans_blocks(inode
);
3581 credits
= ext4_blocks_for_truncate(inode
);
3583 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3584 if (IS_ERR(handle
)) {
3585 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
3589 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
3590 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
3593 * We add the inode to the orphan list, so that if this
3594 * truncate spans multiple transactions, and we crash, we will
3595 * resume the truncate when the filesystem recovers. It also
3596 * marks the inode dirty, to catch the new size.
3598 * Implication: the file must always be in a sane, consistent
3599 * truncatable state while each transaction commits.
3601 if (ext4_orphan_add(handle
, inode
))
3604 down_write(&EXT4_I(inode
)->i_data_sem
);
3606 ext4_discard_preallocations(inode
);
3608 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3609 ext4_ext_truncate(handle
, inode
);
3611 ext4_ind_truncate(handle
, inode
);
3613 up_write(&ei
->i_data_sem
);
3616 ext4_handle_sync(handle
);
3620 * If this was a simple ftruncate() and the file will remain alive,
3621 * then we need to clear up the orphan record which we created above.
3622 * However, if this was a real unlink then we were called by
3623 * ext4_evict_inode(), and we allow that function to clean up the
3624 * orphan info for us.
3627 ext4_orphan_del(handle
, inode
);
3629 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3630 ext4_mark_inode_dirty(handle
, inode
);
3631 ext4_journal_stop(handle
);
3633 trace_ext4_truncate_exit(inode
);
3637 * ext4_get_inode_loc returns with an extra refcount against the inode's
3638 * underlying buffer_head on success. If 'in_mem' is true, we have all
3639 * data in memory that is needed to recreate the on-disk version of this
3642 static int __ext4_get_inode_loc(struct inode
*inode
,
3643 struct ext4_iloc
*iloc
, int in_mem
)
3645 struct ext4_group_desc
*gdp
;
3646 struct buffer_head
*bh
;
3647 struct super_block
*sb
= inode
->i_sb
;
3649 int inodes_per_block
, inode_offset
;
3652 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3655 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3656 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3661 * Figure out the offset within the block group inode table
3663 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3664 inode_offset
= ((inode
->i_ino
- 1) %
3665 EXT4_INODES_PER_GROUP(sb
));
3666 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3667 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3669 bh
= sb_getblk(sb
, block
);
3672 if (!buffer_uptodate(bh
)) {
3676 * If the buffer has the write error flag, we have failed
3677 * to write out another inode in the same block. In this
3678 * case, we don't have to read the block because we may
3679 * read the old inode data successfully.
3681 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3682 set_buffer_uptodate(bh
);
3684 if (buffer_uptodate(bh
)) {
3685 /* someone brought it uptodate while we waited */
3691 * If we have all information of the inode in memory and this
3692 * is the only valid inode in the block, we need not read the
3696 struct buffer_head
*bitmap_bh
;
3699 start
= inode_offset
& ~(inodes_per_block
- 1);
3701 /* Is the inode bitmap in cache? */
3702 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3703 if (unlikely(!bitmap_bh
))
3707 * If the inode bitmap isn't in cache then the
3708 * optimisation may end up performing two reads instead
3709 * of one, so skip it.
3711 if (!buffer_uptodate(bitmap_bh
)) {
3715 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3716 if (i
== inode_offset
)
3718 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3722 if (i
== start
+ inodes_per_block
) {
3723 /* all other inodes are free, so skip I/O */
3724 memset(bh
->b_data
, 0, bh
->b_size
);
3725 set_buffer_uptodate(bh
);
3733 * If we need to do any I/O, try to pre-readahead extra
3734 * blocks from the inode table.
3736 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3737 ext4_fsblk_t b
, end
, table
;
3739 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
3741 table
= ext4_inode_table(sb
, gdp
);
3742 /* s_inode_readahead_blks is always a power of 2 */
3743 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
3747 num
= EXT4_INODES_PER_GROUP(sb
);
3748 if (ext4_has_group_desc_csum(sb
))
3749 num
-= ext4_itable_unused_count(sb
, gdp
);
3750 table
+= num
/ inodes_per_block
;
3754 sb_breadahead(sb
, b
++);
3758 * There are other valid inodes in the buffer, this inode
3759 * has in-inode xattrs, or we don't have this inode in memory.
3760 * Read the block from disk.
3762 trace_ext4_load_inode(inode
);
3764 bh
->b_end_io
= end_buffer_read_sync
;
3765 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
3767 if (!buffer_uptodate(bh
)) {
3768 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3769 "unable to read itable block");
3779 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3781 /* We have all inode data except xattrs in memory here. */
3782 return __ext4_get_inode_loc(inode
, iloc
,
3783 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3786 void ext4_set_inode_flags(struct inode
*inode
)
3788 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3789 unsigned int new_fl
= 0;
3791 if (flags
& EXT4_SYNC_FL
)
3793 if (flags
& EXT4_APPEND_FL
)
3795 if (flags
& EXT4_IMMUTABLE_FL
)
3796 new_fl
|= S_IMMUTABLE
;
3797 if (flags
& EXT4_NOATIME_FL
)
3798 new_fl
|= S_NOATIME
;
3799 if (flags
& EXT4_DIRSYNC_FL
)
3800 new_fl
|= S_DIRSYNC
;
3801 inode_set_flags(inode
, new_fl
,
3802 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3805 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3806 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3808 unsigned int vfs_fl
;
3809 unsigned long old_fl
, new_fl
;
3812 vfs_fl
= ei
->vfs_inode
.i_flags
;
3813 old_fl
= ei
->i_flags
;
3814 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3815 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3817 if (vfs_fl
& S_SYNC
)
3818 new_fl
|= EXT4_SYNC_FL
;
3819 if (vfs_fl
& S_APPEND
)
3820 new_fl
|= EXT4_APPEND_FL
;
3821 if (vfs_fl
& S_IMMUTABLE
)
3822 new_fl
|= EXT4_IMMUTABLE_FL
;
3823 if (vfs_fl
& S_NOATIME
)
3824 new_fl
|= EXT4_NOATIME_FL
;
3825 if (vfs_fl
& S_DIRSYNC
)
3826 new_fl
|= EXT4_DIRSYNC_FL
;
3827 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3830 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3831 struct ext4_inode_info
*ei
)
3834 struct inode
*inode
= &(ei
->vfs_inode
);
3835 struct super_block
*sb
= inode
->i_sb
;
3837 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3838 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3839 /* we are using combined 48 bit field */
3840 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3841 le32_to_cpu(raw_inode
->i_blocks_lo
);
3842 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3843 /* i_blocks represent file system block size */
3844 return i_blocks
<< (inode
->i_blkbits
- 9);
3849 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3853 static inline void ext4_iget_extra_inode(struct inode
*inode
,
3854 struct ext4_inode
*raw_inode
,
3855 struct ext4_inode_info
*ei
)
3857 __le32
*magic
= (void *)raw_inode
+
3858 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
3859 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
3860 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
3861 ext4_find_inline_data_nolock(inode
);
3863 EXT4_I(inode
)->i_inline_off
= 0;
3866 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3868 struct ext4_iloc iloc
;
3869 struct ext4_inode
*raw_inode
;
3870 struct ext4_inode_info
*ei
;
3871 struct inode
*inode
;
3872 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
3878 inode
= iget_locked(sb
, ino
);
3880 return ERR_PTR(-ENOMEM
);
3881 if (!(inode
->i_state
& I_NEW
))
3887 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3890 raw_inode
= ext4_raw_inode(&iloc
);
3892 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3893 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3894 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3895 EXT4_INODE_SIZE(inode
->i_sb
)) {
3896 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
3897 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
3898 EXT4_INODE_SIZE(inode
->i_sb
));
3903 ei
->i_extra_isize
= 0;
3905 /* Precompute checksum seed for inode metadata */
3906 if (ext4_has_metadata_csum(sb
)) {
3907 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3909 __le32 inum
= cpu_to_le32(inode
->i_ino
);
3910 __le32 gen
= raw_inode
->i_generation
;
3911 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
3913 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
3917 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
3918 EXT4_ERROR_INODE(inode
, "checksum invalid");
3923 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3924 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3925 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3926 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3927 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3928 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3930 i_uid_write(inode
, i_uid
);
3931 i_gid_write(inode
, i_gid
);
3932 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
3934 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
3935 ei
->i_inline_off
= 0;
3936 ei
->i_dir_start_lookup
= 0;
3937 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3938 /* We now have enough fields to check if the inode was active or not.
3939 * This is needed because nfsd might try to access dead inodes
3940 * the test is that same one that e2fsck uses
3941 * NeilBrown 1999oct15
3943 if (inode
->i_nlink
== 0) {
3944 if ((inode
->i_mode
== 0 ||
3945 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
3946 ino
!= EXT4_BOOT_LOADER_INO
) {
3947 /* this inode is deleted */
3951 /* The only unlinked inodes we let through here have
3952 * valid i_mode and are being read by the orphan
3953 * recovery code: that's fine, we're about to complete
3954 * the process of deleting those.
3955 * OR it is the EXT4_BOOT_LOADER_INO which is
3956 * not initialized on a new filesystem. */
3958 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3959 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3960 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3961 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
3963 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3964 inode
->i_size
= ext4_isize(raw_inode
);
3965 ei
->i_disksize
= inode
->i_size
;
3967 ei
->i_reserved_quota
= 0;
3969 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3970 ei
->i_block_group
= iloc
.block_group
;
3971 ei
->i_last_alloc_group
= ~0;
3973 * NOTE! The in-memory inode i_data array is in little-endian order
3974 * even on big-endian machines: we do NOT byteswap the block numbers!
3976 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3977 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
3978 INIT_LIST_HEAD(&ei
->i_orphan
);
3981 * Set transaction id's of transactions that have to be committed
3982 * to finish f[data]sync. We set them to currently running transaction
3983 * as we cannot be sure that the inode or some of its metadata isn't
3984 * part of the transaction - the inode could have been reclaimed and
3985 * now it is reread from disk.
3988 transaction_t
*transaction
;
3991 read_lock(&journal
->j_state_lock
);
3992 if (journal
->j_running_transaction
)
3993 transaction
= journal
->j_running_transaction
;
3995 transaction
= journal
->j_committing_transaction
;
3997 tid
= transaction
->t_tid
;
3999 tid
= journal
->j_commit_sequence
;
4000 read_unlock(&journal
->j_state_lock
);
4001 ei
->i_sync_tid
= tid
;
4002 ei
->i_datasync_tid
= tid
;
4005 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4006 if (ei
->i_extra_isize
== 0) {
4007 /* The extra space is currently unused. Use it. */
4008 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4009 EXT4_GOOD_OLD_INODE_SIZE
;
4011 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4015 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4016 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4017 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4018 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4020 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4021 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4022 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4023 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4025 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4030 if (ei
->i_file_acl
&&
4031 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4032 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4036 } else if (!ext4_has_inline_data(inode
)) {
4037 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4038 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4039 (S_ISLNK(inode
->i_mode
) &&
4040 !ext4_inode_is_fast_symlink(inode
))))
4041 /* Validate extent which is part of inode */
4042 ret
= ext4_ext_check_inode(inode
);
4043 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4044 (S_ISLNK(inode
->i_mode
) &&
4045 !ext4_inode_is_fast_symlink(inode
))) {
4046 /* Validate block references which are part of inode */
4047 ret
= ext4_ind_check_inode(inode
);
4053 if (S_ISREG(inode
->i_mode
)) {
4054 inode
->i_op
= &ext4_file_inode_operations
;
4055 inode
->i_fop
= &ext4_file_operations
;
4056 ext4_set_aops(inode
);
4057 } else if (S_ISDIR(inode
->i_mode
)) {
4058 inode
->i_op
= &ext4_dir_inode_operations
;
4059 inode
->i_fop
= &ext4_dir_operations
;
4060 } else if (S_ISLNK(inode
->i_mode
)) {
4061 if (ext4_inode_is_fast_symlink(inode
)) {
4062 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4063 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4064 sizeof(ei
->i_data
) - 1);
4066 inode
->i_op
= &ext4_symlink_inode_operations
;
4067 ext4_set_aops(inode
);
4069 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4070 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4071 inode
->i_op
= &ext4_special_inode_operations
;
4072 if (raw_inode
->i_block
[0])
4073 init_special_inode(inode
, inode
->i_mode
,
4074 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4076 init_special_inode(inode
, inode
->i_mode
,
4077 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4078 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4079 make_bad_inode(inode
);
4082 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4086 ext4_set_inode_flags(inode
);
4087 unlock_new_inode(inode
);
4093 return ERR_PTR(ret
);
4096 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4098 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4099 return ERR_PTR(-EIO
);
4100 return ext4_iget(sb
, ino
);
4103 static int ext4_inode_blocks_set(handle_t
*handle
,
4104 struct ext4_inode
*raw_inode
,
4105 struct ext4_inode_info
*ei
)
4107 struct inode
*inode
= &(ei
->vfs_inode
);
4108 u64 i_blocks
= inode
->i_blocks
;
4109 struct super_block
*sb
= inode
->i_sb
;
4111 if (i_blocks
<= ~0U) {
4113 * i_blocks can be represented in a 32 bit variable
4114 * as multiple of 512 bytes
4116 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4117 raw_inode
->i_blocks_high
= 0;
4118 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4121 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4124 if (i_blocks
<= 0xffffffffffffULL
) {
4126 * i_blocks can be represented in a 48 bit variable
4127 * as multiple of 512 bytes
4129 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4130 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4131 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4133 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4134 /* i_block is stored in file system block size */
4135 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4136 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4137 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4143 * Post the struct inode info into an on-disk inode location in the
4144 * buffer-cache. This gobbles the caller's reference to the
4145 * buffer_head in the inode location struct.
4147 * The caller must have write access to iloc->bh.
4149 static int ext4_do_update_inode(handle_t
*handle
,
4150 struct inode
*inode
,
4151 struct ext4_iloc
*iloc
)
4153 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4154 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4155 struct buffer_head
*bh
= iloc
->bh
;
4156 struct super_block
*sb
= inode
->i_sb
;
4157 int err
= 0, rc
, block
;
4158 int need_datasync
= 0, set_large_file
= 0;
4162 spin_lock(&ei
->i_raw_lock
);
4164 /* For fields not tracked in the in-memory inode,
4165 * initialise them to zero for new inodes. */
4166 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4167 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4169 ext4_get_inode_flags(ei
);
4170 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4171 i_uid
= i_uid_read(inode
);
4172 i_gid
= i_gid_read(inode
);
4173 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4174 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4175 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4177 * Fix up interoperability with old kernels. Otherwise, old inodes get
4178 * re-used with the upper 16 bits of the uid/gid intact
4181 raw_inode
->i_uid_high
=
4182 cpu_to_le16(high_16_bits(i_uid
));
4183 raw_inode
->i_gid_high
=
4184 cpu_to_le16(high_16_bits(i_gid
));
4186 raw_inode
->i_uid_high
= 0;
4187 raw_inode
->i_gid_high
= 0;
4190 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4191 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4192 raw_inode
->i_uid_high
= 0;
4193 raw_inode
->i_gid_high
= 0;
4195 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4197 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4198 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4199 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4200 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4202 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
4204 spin_unlock(&ei
->i_raw_lock
);
4207 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4208 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4209 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4210 raw_inode
->i_file_acl_high
=
4211 cpu_to_le16(ei
->i_file_acl
>> 32);
4212 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4213 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4214 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4217 if (ei
->i_disksize
> 0x7fffffffULL
) {
4218 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4219 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4220 EXT4_SB(sb
)->s_es
->s_rev_level
==
4221 cpu_to_le32(EXT4_GOOD_OLD_REV
))
4224 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4225 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4226 if (old_valid_dev(inode
->i_rdev
)) {
4227 raw_inode
->i_block
[0] =
4228 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4229 raw_inode
->i_block
[1] = 0;
4231 raw_inode
->i_block
[0] = 0;
4232 raw_inode
->i_block
[1] =
4233 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4234 raw_inode
->i_block
[2] = 0;
4236 } else if (!ext4_has_inline_data(inode
)) {
4237 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4238 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4241 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4242 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4243 if (ei
->i_extra_isize
) {
4244 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4245 raw_inode
->i_version_hi
=
4246 cpu_to_le32(inode
->i_version
>> 32);
4247 raw_inode
->i_extra_isize
=
4248 cpu_to_le16(ei
->i_extra_isize
);
4252 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4254 spin_unlock(&ei
->i_raw_lock
);
4256 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4257 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4260 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4261 if (set_large_file
) {
4262 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
4263 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
4266 ext4_update_dynamic_rev(sb
);
4267 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4268 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4269 ext4_handle_sync(handle
);
4270 err
= ext4_handle_dirty_super(handle
, sb
);
4272 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4275 ext4_std_error(inode
->i_sb
, err
);
4280 * ext4_write_inode()
4282 * We are called from a few places:
4284 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4285 * Here, there will be no transaction running. We wait for any running
4286 * transaction to commit.
4288 * - Within flush work (sys_sync(), kupdate and such).
4289 * We wait on commit, if told to.
4291 * - Within iput_final() -> write_inode_now()
4292 * We wait on commit, if told to.
4294 * In all cases it is actually safe for us to return without doing anything,
4295 * because the inode has been copied into a raw inode buffer in
4296 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4299 * Note that we are absolutely dependent upon all inode dirtiers doing the
4300 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4301 * which we are interested.
4303 * It would be a bug for them to not do this. The code:
4305 * mark_inode_dirty(inode)
4307 * inode->i_size = expr;
4309 * is in error because write_inode() could occur while `stuff()' is running,
4310 * and the new i_size will be lost. Plus the inode will no longer be on the
4311 * superblock's dirty inode list.
4313 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4317 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
4320 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4321 if (ext4_journal_current_handle()) {
4322 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4328 * No need to force transaction in WB_SYNC_NONE mode. Also
4329 * ext4_sync_fs() will force the commit after everything is
4332 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
4335 err
= ext4_force_commit(inode
->i_sb
);
4337 struct ext4_iloc iloc
;
4339 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4343 * sync(2) will flush the whole buffer cache. No need to do
4344 * it here separately for each inode.
4346 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
4347 sync_dirty_buffer(iloc
.bh
);
4348 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4349 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4350 "IO error syncing inode");
4359 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4360 * buffers that are attached to a page stradding i_size and are undergoing
4361 * commit. In that case we have to wait for commit to finish and try again.
4363 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4367 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4368 tid_t commit_tid
= 0;
4371 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4373 * All buffers in the last page remain valid? Then there's nothing to
4374 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4377 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4380 page
= find_lock_page(inode
->i_mapping
,
4381 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4384 ret
= __ext4_journalled_invalidatepage(page
, offset
,
4385 PAGE_CACHE_SIZE
- offset
);
4387 page_cache_release(page
);
4391 read_lock(&journal
->j_state_lock
);
4392 if (journal
->j_committing_transaction
)
4393 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4394 read_unlock(&journal
->j_state_lock
);
4396 jbd2_log_wait_commit(journal
, commit_tid
);
4403 * Called from notify_change.
4405 * We want to trap VFS attempts to truncate the file as soon as
4406 * possible. In particular, we want to make sure that when the VFS
4407 * shrinks i_size, we put the inode on the orphan list and modify
4408 * i_disksize immediately, so that during the subsequent flushing of
4409 * dirty pages and freeing of disk blocks, we can guarantee that any
4410 * commit will leave the blocks being flushed in an unused state on
4411 * disk. (On recovery, the inode will get truncated and the blocks will
4412 * be freed, so we have a strong guarantee that no future commit will
4413 * leave these blocks visible to the user.)
4415 * Another thing we have to assure is that if we are in ordered mode
4416 * and inode is still attached to the committing transaction, we must
4417 * we start writeout of all the dirty pages which are being truncated.
4418 * This way we are sure that all the data written in the previous
4419 * transaction are already on disk (truncate waits for pages under
4422 * Called with inode->i_mutex down.
4424 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4426 struct inode
*inode
= dentry
->d_inode
;
4429 const unsigned int ia_valid
= attr
->ia_valid
;
4431 error
= inode_change_ok(inode
, attr
);
4435 if (is_quota_modification(inode
, attr
))
4436 dquot_initialize(inode
);
4437 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4438 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4441 /* (user+group)*(old+new) structure, inode write (sb,
4442 * inode block, ? - but truncate inode update has it) */
4443 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4444 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4445 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4446 if (IS_ERR(handle
)) {
4447 error
= PTR_ERR(handle
);
4450 error
= dquot_transfer(inode
, attr
);
4452 ext4_journal_stop(handle
);
4455 /* Update corresponding info in inode so that everything is in
4456 * one transaction */
4457 if (attr
->ia_valid
& ATTR_UID
)
4458 inode
->i_uid
= attr
->ia_uid
;
4459 if (attr
->ia_valid
& ATTR_GID
)
4460 inode
->i_gid
= attr
->ia_gid
;
4461 error
= ext4_mark_inode_dirty(handle
, inode
);
4462 ext4_journal_stop(handle
);
4465 if (attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
!= inode
->i_size
) {
4468 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4469 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4471 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4475 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
4476 inode_inc_iversion(inode
);
4478 if (S_ISREG(inode
->i_mode
) &&
4479 (attr
->ia_size
< inode
->i_size
)) {
4480 if (ext4_should_order_data(inode
)) {
4481 error
= ext4_begin_ordered_truncate(inode
,
4486 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4487 if (IS_ERR(handle
)) {
4488 error
= PTR_ERR(handle
);
4491 if (ext4_handle_valid(handle
)) {
4492 error
= ext4_orphan_add(handle
, inode
);
4495 down_write(&EXT4_I(inode
)->i_data_sem
);
4496 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4497 rc
= ext4_mark_inode_dirty(handle
, inode
);
4501 * We have to update i_size under i_data_sem together
4502 * with i_disksize to avoid races with writeback code
4503 * running ext4_wb_update_i_disksize().
4506 i_size_write(inode
, attr
->ia_size
);
4507 up_write(&EXT4_I(inode
)->i_data_sem
);
4508 ext4_journal_stop(handle
);
4510 ext4_orphan_del(NULL
, inode
);
4514 loff_t oldsize
= inode
->i_size
;
4516 i_size_write(inode
, attr
->ia_size
);
4517 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
4521 * Blocks are going to be removed from the inode. Wait
4522 * for dio in flight. Temporarily disable
4523 * dioread_nolock to prevent livelock.
4526 if (!ext4_should_journal_data(inode
)) {
4527 ext4_inode_block_unlocked_dio(inode
);
4528 inode_dio_wait(inode
);
4529 ext4_inode_resume_unlocked_dio(inode
);
4531 ext4_wait_for_tail_page_commit(inode
);
4534 * Truncate pagecache after we've waited for commit
4535 * in data=journal mode to make pages freeable.
4537 truncate_pagecache(inode
, inode
->i_size
);
4540 * We want to call ext4_truncate() even if attr->ia_size ==
4541 * inode->i_size for cases like truncation of fallocated space
4543 if (attr
->ia_valid
& ATTR_SIZE
)
4544 ext4_truncate(inode
);
4547 setattr_copy(inode
, attr
);
4548 mark_inode_dirty(inode
);
4552 * If the call to ext4_truncate failed to get a transaction handle at
4553 * all, we need to clean up the in-core orphan list manually.
4555 if (orphan
&& inode
->i_nlink
)
4556 ext4_orphan_del(NULL
, inode
);
4558 if (!rc
&& (ia_valid
& ATTR_MODE
))
4559 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
4562 ext4_std_error(inode
->i_sb
, error
);
4568 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4571 struct inode
*inode
;
4572 unsigned long long delalloc_blocks
;
4574 inode
= dentry
->d_inode
;
4575 generic_fillattr(inode
, stat
);
4578 * If there is inline data in the inode, the inode will normally not
4579 * have data blocks allocated (it may have an external xattr block).
4580 * Report at least one sector for such files, so tools like tar, rsync,
4581 * others doen't incorrectly think the file is completely sparse.
4583 if (unlikely(ext4_has_inline_data(inode
)))
4584 stat
->blocks
+= (stat
->size
+ 511) >> 9;
4587 * We can't update i_blocks if the block allocation is delayed
4588 * otherwise in the case of system crash before the real block
4589 * allocation is done, we will have i_blocks inconsistent with
4590 * on-disk file blocks.
4591 * We always keep i_blocks updated together with real
4592 * allocation. But to not confuse with user, stat
4593 * will return the blocks that include the delayed allocation
4594 * blocks for this file.
4596 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4597 EXT4_I(inode
)->i_reserved_data_blocks
);
4598 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
4602 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
4605 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4606 return ext4_ind_trans_blocks(inode
, lblocks
);
4607 return ext4_ext_index_trans_blocks(inode
, pextents
);
4611 * Account for index blocks, block groups bitmaps and block group
4612 * descriptor blocks if modify datablocks and index blocks
4613 * worse case, the indexs blocks spread over different block groups
4615 * If datablocks are discontiguous, they are possible to spread over
4616 * different block groups too. If they are contiguous, with flexbg,
4617 * they could still across block group boundary.
4619 * Also account for superblock, inode, quota and xattr blocks
4621 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
4624 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4630 * How many index blocks need to touch to map @lblocks logical blocks
4631 * to @pextents physical extents?
4633 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
4638 * Now let's see how many group bitmaps and group descriptors need
4641 groups
= idxblocks
+ pextents
;
4643 if (groups
> ngroups
)
4645 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4646 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4648 /* bitmaps and block group descriptor blocks */
4649 ret
+= groups
+ gdpblocks
;
4651 /* Blocks for super block, inode, quota and xattr blocks */
4652 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4658 * Calculate the total number of credits to reserve to fit
4659 * the modification of a single pages into a single transaction,
4660 * which may include multiple chunks of block allocations.
4662 * This could be called via ext4_write_begin()
4664 * We need to consider the worse case, when
4665 * one new block per extent.
4667 int ext4_writepage_trans_blocks(struct inode
*inode
)
4669 int bpp
= ext4_journal_blocks_per_page(inode
);
4672 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
4674 /* Account for data blocks for journalled mode */
4675 if (ext4_should_journal_data(inode
))
4681 * Calculate the journal credits for a chunk of data modification.
4683 * This is called from DIO, fallocate or whoever calling
4684 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4686 * journal buffers for data blocks are not included here, as DIO
4687 * and fallocate do no need to journal data buffers.
4689 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4691 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4695 * The caller must have previously called ext4_reserve_inode_write().
4696 * Give this, we know that the caller already has write access to iloc->bh.
4698 int ext4_mark_iloc_dirty(handle_t
*handle
,
4699 struct inode
*inode
, struct ext4_iloc
*iloc
)
4703 if (IS_I_VERSION(inode
))
4704 inode_inc_iversion(inode
);
4706 /* the do_update_inode consumes one bh->b_count */
4709 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4710 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4716 * On success, We end up with an outstanding reference count against
4717 * iloc->bh. This _must_ be cleaned up later.
4721 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4722 struct ext4_iloc
*iloc
)
4726 err
= ext4_get_inode_loc(inode
, iloc
);
4728 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4729 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4735 ext4_std_error(inode
->i_sb
, err
);
4740 * Expand an inode by new_extra_isize bytes.
4741 * Returns 0 on success or negative error number on failure.
4743 static int ext4_expand_extra_isize(struct inode
*inode
,
4744 unsigned int new_extra_isize
,
4745 struct ext4_iloc iloc
,
4748 struct ext4_inode
*raw_inode
;
4749 struct ext4_xattr_ibody_header
*header
;
4751 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4754 raw_inode
= ext4_raw_inode(&iloc
);
4756 header
= IHDR(inode
, raw_inode
);
4758 /* No extended attributes present */
4759 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4760 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4761 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4763 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4767 /* try to expand with EAs present */
4768 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4773 * What we do here is to mark the in-core inode as clean with respect to inode
4774 * dirtiness (it may still be data-dirty).
4775 * This means that the in-core inode may be reaped by prune_icache
4776 * without having to perform any I/O. This is a very good thing,
4777 * because *any* task may call prune_icache - even ones which
4778 * have a transaction open against a different journal.
4780 * Is this cheating? Not really. Sure, we haven't written the
4781 * inode out, but prune_icache isn't a user-visible syncing function.
4782 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4783 * we start and wait on commits.
4785 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4787 struct ext4_iloc iloc
;
4788 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4789 static unsigned int mnt_count
;
4793 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4794 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4795 if (ext4_handle_valid(handle
) &&
4796 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4797 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4799 * We need extra buffer credits since we may write into EA block
4800 * with this same handle. If journal_extend fails, then it will
4801 * only result in a minor loss of functionality for that inode.
4802 * If this is felt to be critical, then e2fsck should be run to
4803 * force a large enough s_min_extra_isize.
4805 if ((jbd2_journal_extend(handle
,
4806 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4807 ret
= ext4_expand_extra_isize(inode
,
4808 sbi
->s_want_extra_isize
,
4811 ext4_set_inode_state(inode
,
4812 EXT4_STATE_NO_EXPAND
);
4814 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4815 ext4_warning(inode
->i_sb
,
4816 "Unable to expand inode %lu. Delete"
4817 " some EAs or run e2fsck.",
4820 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4826 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4831 * ext4_dirty_inode() is called from __mark_inode_dirty()
4833 * We're really interested in the case where a file is being extended.
4834 * i_size has been changed by generic_commit_write() and we thus need
4835 * to include the updated inode in the current transaction.
4837 * Also, dquot_alloc_block() will always dirty the inode when blocks
4838 * are allocated to the file.
4840 * If the inode is marked synchronous, we don't honour that here - doing
4841 * so would cause a commit on atime updates, which we don't bother doing.
4842 * We handle synchronous inodes at the highest possible level.
4844 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4848 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
4852 ext4_mark_inode_dirty(handle
, inode
);
4854 ext4_journal_stop(handle
);
4861 * Bind an inode's backing buffer_head into this transaction, to prevent
4862 * it from being flushed to disk early. Unlike
4863 * ext4_reserve_inode_write, this leaves behind no bh reference and
4864 * returns no iloc structure, so the caller needs to repeat the iloc
4865 * lookup to mark the inode dirty later.
4867 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4869 struct ext4_iloc iloc
;
4873 err
= ext4_get_inode_loc(inode
, &iloc
);
4875 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4876 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4878 err
= ext4_handle_dirty_metadata(handle
,
4884 ext4_std_error(inode
->i_sb
, err
);
4889 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4896 * We have to be very careful here: changing a data block's
4897 * journaling status dynamically is dangerous. If we write a
4898 * data block to the journal, change the status and then delete
4899 * that block, we risk forgetting to revoke the old log record
4900 * from the journal and so a subsequent replay can corrupt data.
4901 * So, first we make sure that the journal is empty and that
4902 * nobody is changing anything.
4905 journal
= EXT4_JOURNAL(inode
);
4908 if (is_journal_aborted(journal
))
4910 /* We have to allocate physical blocks for delalloc blocks
4911 * before flushing journal. otherwise delalloc blocks can not
4912 * be allocated any more. even more truncate on delalloc blocks
4913 * could trigger BUG by flushing delalloc blocks in journal.
4914 * There is no delalloc block in non-journal data mode.
4916 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
4917 err
= ext4_alloc_da_blocks(inode
);
4922 /* Wait for all existing dio workers */
4923 ext4_inode_block_unlocked_dio(inode
);
4924 inode_dio_wait(inode
);
4926 jbd2_journal_lock_updates(journal
);
4929 * OK, there are no updates running now, and all cached data is
4930 * synced to disk. We are now in a completely consistent state
4931 * which doesn't have anything in the journal, and we know that
4932 * no filesystem updates are running, so it is safe to modify
4933 * the inode's in-core data-journaling state flag now.
4937 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4939 err
= jbd2_journal_flush(journal
);
4941 jbd2_journal_unlock_updates(journal
);
4942 ext4_inode_resume_unlocked_dio(inode
);
4945 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4947 ext4_set_aops(inode
);
4949 jbd2_journal_unlock_updates(journal
);
4950 ext4_inode_resume_unlocked_dio(inode
);
4952 /* Finally we can mark the inode as dirty. */
4954 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
4956 return PTR_ERR(handle
);
4958 err
= ext4_mark_inode_dirty(handle
, inode
);
4959 ext4_handle_sync(handle
);
4960 ext4_journal_stop(handle
);
4961 ext4_std_error(inode
->i_sb
, err
);
4966 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4968 return !buffer_mapped(bh
);
4971 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4973 struct page
*page
= vmf
->page
;
4977 struct file
*file
= vma
->vm_file
;
4978 struct inode
*inode
= file_inode(file
);
4979 struct address_space
*mapping
= inode
->i_mapping
;
4981 get_block_t
*get_block
;
4984 sb_start_pagefault(inode
->i_sb
);
4985 file_update_time(vma
->vm_file
);
4986 /* Delalloc case is easy... */
4987 if (test_opt(inode
->i_sb
, DELALLOC
) &&
4988 !ext4_should_journal_data(inode
) &&
4989 !ext4_nonda_switch(inode
->i_sb
)) {
4991 ret
= __block_page_mkwrite(vma
, vmf
,
4992 ext4_da_get_block_prep
);
4993 } while (ret
== -ENOSPC
&&
4994 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
4999 size
= i_size_read(inode
);
5000 /* Page got truncated from under us? */
5001 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5003 ret
= VM_FAULT_NOPAGE
;
5007 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5008 len
= size
& ~PAGE_CACHE_MASK
;
5010 len
= PAGE_CACHE_SIZE
;
5012 * Return if we have all the buffers mapped. This avoids the need to do
5013 * journal_start/journal_stop which can block and take a long time
5015 if (page_has_buffers(page
)) {
5016 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5018 ext4_bh_unmapped
)) {
5019 /* Wait so that we don't change page under IO */
5020 wait_for_stable_page(page
);
5021 ret
= VM_FAULT_LOCKED
;
5026 /* OK, we need to fill the hole... */
5027 if (ext4_should_dioread_nolock(inode
))
5028 get_block
= ext4_get_block_write
;
5030 get_block
= ext4_get_block
;
5032 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5033 ext4_writepage_trans_blocks(inode
));
5034 if (IS_ERR(handle
)) {
5035 ret
= VM_FAULT_SIGBUS
;
5038 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
5039 if (!ret
&& ext4_should_journal_data(inode
)) {
5040 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5041 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5043 ret
= VM_FAULT_SIGBUS
;
5044 ext4_journal_stop(handle
);
5047 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5049 ext4_journal_stop(handle
);
5050 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5053 ret
= block_page_mkwrite_return(ret
);
5055 sb_end_pagefault(inode
->i_sb
);